What are people building with Jev?

jev is INSANE. in 40 seconds it broke down 724 live ads from 37 brands. every hook. every format. offer. cta. awareness stage. landing page mismatch. used 9 cents of tokens. (will be avail in @stealads + mcp) https://t.co/II1T9hy5tg

63.5万 views · 5.5k likes · 7.5k saves

JEV is INSANE. We gave it 400 companies and one candidate profile. In 12 seconds, it predicted which jobs the candidate had the highest chance of getting, assigned a confidence score and detected job-candidate mismatches. All for just $0.0005 It can also score companies, analyse your experience, match you with the right roles and identify the opportunities you’re most likely to get based on your profile. Coming soon to @textbackdoor Comment “JEV” for early access.

3.7万 views · 553 likes · 543 saves

jev is insane 🤯 Here is Jev playing subway surfers at super human speed, and also playing 50 games at once. cost less than a cent to do this run. Jev does not replace llms like astra or fable, but opens up an entirely new world of capabilities. https://t.co/T9QrLKmUcG

24.3万 views · 3.4k likes · 1.6k saves

whoa this actually worked! Jev lets me control my browser in real time with my voice now > i talk > transcript sent to Jev > jev returns probabilities in ~300ms > browser clicks costs: $0.0002 per decision i'm stunned how fast this is. when i asked it to "go back", it even finished the request before i finished my sentence 😂

25.8万 views · 3.3k likes · 2.5k saves

Just trying out Jev, I made a Chrome extension that: - Listens to your YouTube audio (optional) - Detects if it gets to a sponsor segment - Skips it ➡️➡️➡️ - All in real-time while costing ~$0.005 per video Prototype project, BYOK, open-source: https://t.co/g4sPXR5kML https://t.co/nH6c1LhU3C

5.5万 views · 897 likes · 589 saves

instead of generating text, jev from @typesafeai generates structured output this makes it great for classification tasks like model routing, tool selection/search, and guardrails of many forms! it's also ridiculously fast and cheap compared to LLMs doing the same tasks https://t.co/59a8PxBuQ6

1.6万 views · 95 likes · 68 saves

Jev is here. - how is different from an LLM? - how does it work under-the-hood? - what are some real life examples? - when should I use it? I answer all that below: https://t.co/HLfPVp51h6

4.7万 views · 899 likes · 994 saves

Jev v1.13 dodges rockets with probability calculation 🚀 @typesafeai's new non-LLM model returns decisions instead of text so we had it calculate a safe tile every 330 ms while rockets fell and it survived 25 of 26 for under a cent Run Jev via API -> https://t.co/RbcCOIgVkj https://t.co/f2VmOEPFnO

5.1万 views · 503 likes · 263 saves

I got a rubik's cube to solve itself with @typesafeai 's Jev and it solves it like a person does, 94 moves, not the 22 move optimal solution. Jev isn't an LLM, it just answers one question in ~250ms with a probability. so I put the beginner method in code (the one you'd learn on youtube: white cross, corners, middle layer, yellow layer) and at every step Jev just looks at the cube and picks which case it's in. code checks every pick. ~4 seconds of model time total. video is slowed down so you can read it!

20.2万 views · 692 likes · 299 saves

视觉版 Jev 来了 https://t.co/x0kzeE3hdx 仿照着社区的思路，把候选结果映射成固定标签，prefill 完直接采样，限制输出 token，采用多模态的 qwen-3.5-0.8B 4bit 量化部署在本地，16g m4 mbp 顺利运行 叠甲：这只是在 infer 层小小的复现一下 jev 的形式，肯定不如各种成熟推理框架效率高，也肯定不是 jev 的正在原理，但是作为 inference 的小入门仍旧是不错的 demo 如下 200ms 实现《三色货品分拣》

2.9万 views · 293 likes · 330 saves

I rebuilt my job crawler with Jev. The task: start at a company's official homepage, find Careers, and identify jobs that match my profile. Before, with an LLM: ~5 minutes. After, with Jev: just over 20 seconds in my test. Every company organizes its website differently. Jev identifies the Careers entry point, chooses which links to follow, recognizes job pages, and scores each role against my profile. This is where Jev makes sense to me: automation that needs lots of small decisions, with faster responses and lower costs than calling an LLM at each step. Packaged it as a skill: jev-job-hunter. Demo below.

1.4万 views · 35 likes · 45 saves

jav @typesafeai - jestem pod wrażeniem 🤩 Cool. Napisałem kolejne demo. Myślę, że teraz to widać potęgę modelu ... i o co chodzi i widać prawdziwą wartość biznesową. Akt I - X-ray: „jedno zapytanie, wszystkie pytania naraz" - dla przykładowej, krótkiej notatki prasowej to 67 osądów w ~1 s za $0.0002; dla dłuższego RAPORTU może być i 145 osądów w 1,5 s za $0.0005. Skoro nie ma tokenów wyjściowych, to dorzucenie kolejnego pytania kosztuje tylko tokeny samego pytania. Zobaczcie jak można dynamicznie klasyfikować i jednocześnie pytać o wiele wymiarów. Akt I odpowiada na pytanie „ile" osądów naraz i za ile. 🥳😍😍😍 Akt II - Firehose: „AI jako zależność w pętli czasu rzeczywistego" - podłączyłem się do publicznego strumienia SSE Wikimedia (stream_wikimedia_org). Filtruję: angielska Wikipedia, przestrzeń główna, człowiek (nie bot). Dla każdej pobieram prawdziwy diff przez REST compare, który zwraca strukturę zamiast HTML-a. Potem 8 pytań o tę edycję: wandalizm, spam promocyjny, atak na osobę, edycja testowa, usuwanie treści, twierdzenie bez źródła, Score „ile szkody" (0–3) i Choice „co zrobić". 71 edycji w 75 sekund, 1,0 decyzji/s, mediana 306 ms, $0,19 za godzinę oceniania całej Wikipedii. Rozkład: 49 zostaw / 20 do człowieka / 2 cofnij. Akt II to „gdzie", że to może siedzieć w pętli czasu rzeczywistego, w rdzeniu systemu. Akt III - autonomia. TypeSafe nie publikuje benchmarków „poziomu inteligencji". Publikuje prędkość i cenę, czyli dokładnie to, co i tak łatwo zweryfikować. Czy zatem model wie, kiedy ma rację zostaje do zmierzenia użytkownikowi. No to zmierzyłem. 300 zapytań do asystenta głosowego, 150 możliwych intencji, a jedna piąta z nich prosi o coś, czego asystent w ogóle nie obsługuje. Jeden przebieg. Przy progu 0.90 z dodatkowym pytaniem o zakres: - 62% ruchu obsługuje się samo, bez człowieka - 99% z tego jest poprawne -> konkretnie 2 błędne odpowiedzi na 186 - 100% zapytań spoza zakresu zatrzymanych - pozostałe 114 ląduje w kolejce do człowieka Wykres po prawej jest ważniejszy niż te liczby. Oś X jakie prawdopodobieństwo model zadeklarował. Oś Y jak często faktycznie miał rację. Przekątna to uczciwość. Gdy mówił 0.99 trafiał w 99%. Rzecz, o której się nie mówi: sama pewność NIE wyłapuje zapytań spoza zakresu. Choice jest relatywny zawsze musi kogoś wskazać, więc wysoka pewność znaczy „ta opcja wygrała z resztą", a nie „ta opcja jest dobra". Dane: CLINC150, publiczny zbiór. Całość: ~20 sekund i 4 centy. Akt III na „czy można na tym polegać" i jak to działa (pomiar jakości).

6.7k views · 84 likes · 83 saves

Doc-OCR router using Jev @typesafeai A Jev-powered router that looks at a PDF page by page, decides which pages actually need OCR, extracts the rest locally. Result: save cost on # OCR pages + speed https://t.co/ZjXqHSjSGh

3.3k views · 66 likes · 88 saves

First Jev use case published live on Bannerbear! Instant field mapping between template and source, when names are slightly different eg template - photo - name - company_name data source (eg Airtable) - avatar - full_name - business Jev figures it out in one click. https://t.co/2fxRQJ2qEh

1.2万 views · 118 likes · 94 saves

built `jev-review` @typesafeai it's an experimental, local-first MCP plugin that gives coding agents a score quality feedback loop across different metrics. agents call jev while they work, get scored, make improvements, and repeat the loop try below 👇 https://t.co/qHP5JSUSvw

4.4万 views · 490 likes · 568 saves

i used Jev to classify a youtubers last 100 videos based on how likely it is to sell me something... it analyzed & assigned a sales intent score to each video within 12 seconds & cost about $0.02 i haven't done a deep dive into how accurate the score is yet, but from a quick glance it looks like a super promising classifier model

1.2万 views · 183 likes · 152 saves

I made @typesafeai 's new ultra fast model, Jev, generate text, even though it shouldn't, that's fine because I can't read, and it can't write** ** up to 20 words for 0.5$, what a steal https://t.co/fWmGQhefTw

6.1万 views · 284 likes · 132 saves

For those trying to understand the difference between Jev and LLMs + how to use it https://t.co/55o5WEq3Va

6.1万 views · 575 likes · 673 saves

I talk. Codex clicks. Watch the browser go⚡ This video is 1× !!! speed. Built Jev Browser: an open-source adapter for your existing Codex browser tools. Say the task naturally; Codex plans, Jev selects, the browser moves. @sama @OpenAI make this native? @DarioAmodei @hackgoofer https://t.co/xzvUPFBbGa

5.7万 views · 103 likes · 144 saves

I gave an evolution simulation to Jev, a small model from @typesafeai that answers typed questions with odds instead of writing text. Two species with opposite DNA. 14 generations. An ice age. 432 creatures. Jev decided who survived, who mated, and what killed each one. https://t.co/Law8fyfy6d

3.1k views · 73 likes · 26 saves

okay so browser use is a legitimate use case agent-browser <> Jev loop goal: wiki race from 'Coffee' to 'Artificial Intelligence' https://t.co/565VqWYgoJ

1.9万 views · 191 likes · 137 saves

Every model wants the job. The fast one says, “I’ve got this.” The reasoning model asks for more context. The safety model says, “Absolutely not.” Jev steps in and sends each task where it belongs. Jev decisions cost $0.04 per 1M input tokens on Aria — with no markup. @CompleteSkeptic The right task. The right model. The right decision.

3.1k views · 21 likes · 0 saves

I increasingly think future software will combine code for control flow with small neural programs for "fuzzy" judgments. That's what I've been exploring with ProgramAsWeights, which answers the question of where those neural programs come from: they are "compiled" from English descriptions. For example, I combined 30 neural programs with a decision tree to build a course website helper that looks like a chatbot but runs locally. Each small program handles a question like "Which specialist answerer should this be routed to", and code controls the overall flow. You can try building with it here: https://t.co/Bwgcn362bc

1.3万 views · 171 likes · 140 saves

Woke up with early access to @typesafeai and was super excited to try. Got a demo cooking before work this morning where typesafe drives an AI co-op partner in my @threejs action game. Can see the log on the right tracking the AI decision making in realtime. Typesafe is reading game state and choosing intent for the bot: When to engage, regroup, survive, or escape. Obviously a bit primitive right now but this is like ~1hr worth of work. I need to run to work, but holy shit super excited to build on this 😄 The total spend for about 10mins play testing this morning was about $0.02. Let's fucking go 😂

6.0k views · 27 likes · 8 saves

found the perfect use case for @typesafeai Jev: instant compaction in 2026, why is compaction still a summarization prompt? Jev can make it instant by scoring every tool call and dropping what’s irrelevant https://t.co/h6NzKzNpCg

265.9万 views · 9.2k likes · 6.8k saves

JEV is INSANE. We gave it 700 high-intent leads and personalised outreach messages. In 40 seconds, it predicted how each message would perform, assigned a confidence score and detected lead-message mismatches. All for just $0.09. JEV can also score leads, analyse buying signals, match each prospect with the best message and identify the campaigns most likely to perform based on data. Coming soon to @GojiberryAI+ MCP. Comment “JEV” for early access.

22.8万 views · 2.1k likes · 2.7k saves

thanks to @typesafeai's Jev we now have massively parallel browser-based adversarial testing suite that tries to break each release. and it costs pennies. https://t.co/WoIlqGjjNB

17.8万 views · 3.6k likes · 2.8k saves

Jev by @typesafeai is now on OpenRouter, in beta. Jev is a System One model. Instead of generating text, it takes your app's state plus a typed question and returns a typed decision with a probability attached. There is no JSON prompting, parsing layer, and nothing to validate against.

41.5万 views · 3.1k likes · 1.4k saves

Built a tax document classifier with Jev. We ingest thousands of tax documents using an LLM pipeline I built last tax season. I read multiple articles as late as April this year claiming AI fails at tax document classification. Wasn't the case back then, and it's proved wrong again now. Jev classifies 100% of our tax document corpus at $0.001 per page. 34x cheaper and 6x faster than the LLM setup. Open sourcing it: https://t.co/gInVHudFRn

4.9万 views · 903 likes · 983 saves

卧槽，居然真有比Codex内置computer use更快的操作电脑的方式 我尝试用Codex+Jev打造了一个加强版computer use，我称之为「Jev Use」。比内置的更快更丝滑，token消耗却差不多 下面是我用「添加Mac日历事件」做了一个对比视频，同样的内容，明显可以看到Jev版本整体过程几乎无任何停顿😆 https://t.co/UCpLInByeE

14.5万 views · 819 likes · 1.0k saves

I trust Jev with my life https://t.co/GN3KpzrvRC

96.1万 views · 4.5k likes · 879 saves

jev is insane 🤯 here is jev playing smash bros against itself he is controlling all 4 different characters. and literally deciding whats the best move to play against itself all within a fraction of a second i used over 22 million tokens to play this match and it only cost me a couple of cents... jev does not replace gpt6 astra but the possibilities with its instant response time are endless

18.7万 views · 2.4k likes · 1.1k saves

Jev unlocks SO many awesome new ideas. I built a macOS app that monitors my Downloads folder along with a customisable set of rules. Is the downloaded file an invoice? Move it to a special folder with the correct filename. No other LLM calls involved - just Jev! https://t.co/iCiIJ9RvlB

7.8万 views · 766 likes · 607 saves

JEV is insane for competitor research! We gave it Resilia's ad library It classified 1,891 ads in 19 seconds for $0.12 costumer journey step + Ad style And a complete deep analysis of the account Coming soon to maxfusion MCP https://t.co/tRfDerHGBs

15.6万 views · 165 likes · 159 saves

Breaking: Browser Use + Jev = Ultrafast ⚡ Findings flights took 7s and cost only $0.0039 🤯 > new action space every step > DOM state space > small LLM fallback to type (this video is at 1x speed btw) Built a tiny open source browser agent. try it below ↓ https://t.co/AplCBRYC5o

248.9万 views · 8.2k likes · 8.9k saves

Full Jev Tutorial What it is, how you can build with it and what new applications it can unlock → 0:00 Intro → 0:34 Jev explained → 4:06 API setup → 5:59 Demo 1: Voice-controlled browser → 11:33 Demo 2: AI memory → 17:27 Demo 3: YouTube predictor https://t.co/V7iK3jZTLZ

11.0万 views · 1.9k likes · 3.5k saves

Another crazy @typesafeai Jev example: Predictive spreadsheets Spreadsheets recalculate numbers, not meaning. Jev reads intent. Type "Urgency" at the top of a column and, as you type, it figures out you want each row rated from "no follow-up needed" to "urgent" in ~100 ms. https://t.co/M6REgq4Q8S

15.8万 views · 982 likes · 733 saves

Jev + SuperX = virality solved ✅ Every post gets 61 questions in ~1s for $0.0004 🤯 > fitted on 9,481 real posts from 207 creators > picks the viral post 2 in 3 times > never rewards reply bait So: write, score, rewrite, stop when it peaks. Free, no signup. try it below ↓ https://t.co/57KJBgkoiv

17.0万 views · 1.1k likes · 1.7k saves

I added Jev to the AI Hedge Fund. We now get frontier-level trading decisions, 100x faster and cheaper than an LLM. How it works: 1 • set strategy 2 • pick tickers 3 • backtest with Jev System now runs in seconds, not minutes.

3.0万 views · 438 likes · 535 saves

Ai is evolving. Jev can be armed at all times. I can speak freely and it knows ( from probabilities ) if im asking my computer to do something or blaberring away at something else. no wake word. speed + affordability + intelligence is getting to the point where an always on ambient jarvis style assistant is possible. Im loving where we are going. CNVS is still the future of vibecoding.

2.2万 views · 560 likes · 427 saves

got Jev to review my PRs. ~200x cheaper than Claude and it answers in half a second 6 real PRs in the video. $0.00007 each. 1,000 PRs = 7 cents vs ~$14.50 on Opus 5 paste a diff → ONE call to @typesafeai → 14 typed checks come back as probabilities: hardcoded secret, sql injection, touches auth, deletes tests, breaks api, migration, debug leftovers, does the description actually match the diff, blast radius, reviewer effort… code turns that into a verdict: BLOCK / security review / nits / merge. anything a critical check isn't sure about (0.35–0.65) gets escalated to a human or a big model instead of guessed

33.1万 views · 2.7k likes · 2.5k saves

i gave jev ₹1,00,000 to trade nifty intraday at 5x leverage. this isn't paper trading but real money on my real kotak account put to use. we hit our ₹1,000 hard stop today after a pretty green morning. will keep tuning the strategy next week and keep y’all posted https://t.co/cCftNveZ31

3.9万 views · 590 likes · 367 saves

jev decided to sacrifice a human to save robots when solving trolley problems https://t.co/vLBwgLYduk

20.7万 views · 1.3k likes · 307 saves

Jev is now in charge of this account's humor https://t.co/ojSOHeMFT2

2.8万 views · 544 likes · 173 saves

e2e + jev from @typesafeai ⚡ I'm building an open-source framework for running e2e tests with agents. supports web, mobile (and more!) available soon: https://t.co/G3XB2L8tiC https://t.co/LvJgAbgvBb

1.7万 views · 351 likes · 313 saves

Okay so Jev can actually do computer use really well Without any screenshots, or LLMs and no Pixels leave my mac I dont even read the Dom elements A local CoreML model segments every button and UI element on screen. On-device OCR reads the labels. That text is all Jev gets. It returns a probability across those elements and tells me the best one to click. Then it clicks, re-runs detection, and decides again. In a loop until the goal is done. ~90ms per decision. Faster than any LLM computer use I've tried. Blazing fast computer use, without any latency @typesafeai is building something really interesting

11.2万 views · 1.6k likes · 1.4k saves

I built a trading bot with Jev! Jev decides if it should "buy" or "sell", given the price feed of an asset pair, and executes real trades. It uses Monad to place the orders on Kuru's on-chain order book in every 300ms block. Demo link → https://t.co/vwl2SUu4jm https://t.co/Sda1G5tXKI

104.6万 views · 4.8k likes · 6.4k saves

Jev will be super helpful for agents to make split second decisions in workflows, data classification, judgment calls, and hundreds of other use-cases in the enterprise. Here's a quick demo with Box and Jev to make that real. The demo pulls an incident report from Box, asks whether it's customer-facing and how severe it is, moves the file into escalate, monitor, or review folders, and sets a metadata template instance with the result. This all happens nearly instantly and at almost no cost. You can imagine this in insurance claims, contract management, loan processing, security reviews, customer log analysis, and so on. Definitely a great new class of AI use-case.

1.7万 views · 167 likes · 151 saves

I built a voice controlled computer-use for my mac using @typesafeai's Jev and it's INSANE how fast it is! I can dictate "open the notes app and create..." and the app opens before I even finish my sentence. https://t.co/n3B7GizUWQ

3.2万 views · 649 likes · 598 saves

ego lite + Jev + DeepSeek Flash = stupid fast. ⚡ 3.71s for 20 Amazon product decisions. GPT-5.6 Sol: 54.45s Same task. Same result. 10/10 on both. Less reasoning. Faster decisions. 1× speed. https://t.co/iJlCVbXSJ8

1.7万 views · 256 likes · 196 saves

Now we're cooking. Got @typesafeai's Jev + AXe controlling the iOS simulator now ultra fast at a fraction of the cost of using an LLM. This is game changing! https://t.co/JK5r23ySoI

9.7万 views · 1.2k likes · 995 saves

卧槽，JEV 的出现很可能就是自动智能交易的起点。 昨天JEV在AI 圈刷屏，今天 Monad 工程师就用JEV做了一个自动交易机器人。 实时读取 MON/USDC 价格，JEV 不负责写分析报告，也不跟你解释一堆逻辑，它只做最核心的事，判断 Buy 还是 Sell，再给出置信度，然后直接把交易发到 Kuru 的链上订单簿执行。 这一下我突然明白 JEV 为什么要做成一个不会说话的模型了。 交易根本不需要 AI 每 300ms 给你写一篇小作文，它需要的是： 行情进来→判断→下单→新行情→重新判断，而且这一整套循环必须足够快、足够便宜。 而 Monad 现在刚好每 300ms 出一个区块，Kuru 又是链上订单簿。高性能链负责执行，JEV 负责决策，两个东西拼起来，已经有点AI 原生交易系统的味道了。 未来的交易 Agent，不一定需要一个会写研报的超级大模型，它更需要一个能在极低延迟下连续做几百万次判断的大脑。 昨天我 提交的WL今天也通过了，接下来我先尝尝咸淡，看看到底怎么个事。

63.5万 views · 3.1k likes · 3.7k saves

Inspired by @typesafeai , here is a Jev-compatible public API to play with It runs a comparable open model (Qwen3.6-35B-A3B), and just uses SGLang radix cache to preserve the prefill reuse / really fast parallel systemone generation - 64 tasks in <1s. https://t.co/bbRh4yAPAB https://t.co/misKQE8n8f

11.0万 views · 873 likes · 821 saves

Jevで次の文章予測判断させるロジック作って、 判断特化AIのJevで、AIチャットできるアプリ作ったんだけど、まず爆速だし、入力は$0.042 / 1Mトークンだけど、出力が無制限で無料でした。 出力無料のAIチャットってなんだこれw 「え？これ作れちゃっていいん？」ってなってる。。。 #Jev https://t.co/h5LDG2tPpN

14.8万 views · 1.1k likes · 1.0k saves

After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI? I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev • 20-200x faster • 40-400x cheaper (w/ output tokens free) • Frontier composable intelligence optimized for decisions AFAICT the shortest path to AI-based economic revolution

3460.4万 views · 6.9万 likes · 5.6万 saves

昨天出的 https://t.co/6eiMFaJTrj Jev 模型 适合高频决策，那岂不又是适合打游戏？ 之前用 GPT-6 Astra 代打杀戮尖塔2 能力强但速度慢。 我刚实测用 Jev 打，行动思考只需 0.7秒， 画面我都没看清它就操作完了。 这超人类游戏速度， 着实又让我震惊瘫坐了😅 语言很难描述这感觉，看视频⬇️ https://t.co/y9Nu78Hnv4

22.5万 views · 1.1k likes · 771 saves

Jev scoring 3000 kid snacks with multiple criteria on https://t.co/KSiodPdZHT in 28 seconds for $0.11 🤯 https://t.co/c9O6uTyiUK

1.6万 views · 132 likes · 103 saves

I rebuilt Tesla Full Self Driving with Jev in less than an hour. This model is a total unlock. https://t.co/mUrvaFrlQ7

55.6万 views · 4.5k likes · 1.8k saves

Jevで、会話をリアルタイムに観測するセンサーを試作しました。 発言のたびに ・探索 ↔ 収束 ・解消 ↔ 未決 ・未合意 ↔ 合意 を更新し、会議の状況を可視化します ユーザーインタビューのリアルタイム解析などにも使えそう。 https://t.co/cO21f4AK99

5.1万 views · 366 likes · 304 saves

TypeGPU + ruNNtime + Jev @typesafeai is a very fun combo :D ruNNtime gives me efficient local inference, TypeGPU lets inference and rendering share GPU resources directly with zero copy. That’s 3 separate NN inferences plus rendering, all happening in realtime Since we control the pipeline, Jev can just sit in the middle and add the semantic bit. camera + mic → Moonshine + YOLO26 + DepthART → Jev → lights, shadows and bloom

7.1万 views · 662 likes · 489 saves

jevデモ② 発話内容に応じたスライドの自動表示。 喋った内容にマッチしたスライドを自動で表示してくれます。PCやポインターを触らずにページを表示できるので、これで誰でもジョブズになれるかも？？ 速い・安いだけでできることがこれだけ広がるのかと感動している。 https://t.co/3UO6DQcbJ2

4.7万 views · 343 likes · 254 saves

Here's a 45-second TL;DR on Jev. I find the core idea beautifully simple, but the video made it really hard to understand. Hope you find it helpful. https://t.co/KtSKlEMVL3

140.9万 views · 8.7k likes · 7.6k saves

"jev can't generate text" But jev has a lot to say! I gave him a couple hundred common english words to choose from + punctuation https://t.co/WmDkTyiG9u

32.8万 views · 3.4k likes · 708 saves

話題のJevで架空のペルソナ150人に製品の導入意向を聞く仕組みを作ってみた。150人ごとに12個の質問をAPIで投げる。 実際にかかった費用は、1.8円でかかった時間は約5秒。十分爆速だけど日本↔︎アメリカのレイテンシが含まれているので本来はもっと早いはず https://t.co/6lbFXY9jjH

23.1万 views · 1.0k likes · 875 saves

Jev + Kimi K3 for fraud detection! TLDR: Jev classified 100 emails in 1.42 seconds, then I routed the uncertain cases to Kimi K3. The full pipeline got 96/100 correct for only ~$0.07. Video is not sped up, check out the live run! Here was my process: I gave Jev 100 emails to classify (a mix of 50 legit & 50 fraudelent emails). It classified all of them in 1.42 seconds. An underrated feature about Jev is it will give you the confidence score for a classification, so I routed any prediction under 95% confidence to Kimi K3 to be fully sure. 31 emails fell below that threshold. After routing those to Kimi K3, the combined pipeline reached 96% accuracy. The full run took 16 seconds & ~$0.07 in inference costs: - $0.068 from Kimi K3 on @togethercompute - $0.003 (1/3 of a cent) from Jev on @typesafeai. I think this is a really interesting pattern: use a fast specialized model like Jev for the narrow task, then route the uncertain cases to a larger LLM. I feel like this kind of approach could be a game changer for use cases like fraud or anything realtime. You can use the speed & low cost of Jev while having a larger LLM as a fallback to ensure high accuracy.

5.3万 views · 843 likes · 739 saves

Jev VS Fable 5.1 VS GPT-6 Astra: chess typesafe's new Jev V13 isn't an LLM. it doesn't chat, doesn't explain, doesn't write code — it only makes decisions. so we made it play blitz against frontier LLMs. the test: 5+0 blitz. every move is one API call. Jev V13 vs Fable 5.1: • fable outplayed it. by move 29 it was +16 in material and even promoted a second queen • but it kept burning 6-15 seconds per move on analysis. jev answered in ~2.6s • so fable didn't have enough time and lost Jev V13 vs Astra: • astra didn't bother winning on material. it mated jev in 18 moves. Qe1#, with 2:27 to spare

52.6万 views · 2.2k likes · 749 saves

If you blink you might miss it Jev + Computer use = 100x faster than any LLM @typesafeai https://t.co/JG09Wx7UCo

8.6万 views · 1.2k likes · 830 saves

I used Jev to classify 1,018 AI research papers. The result: $0.08 total cost and 256ms median end-to-end latency per paper. The pipeline was: 1. Summarize each paper with DeepSeek V4 Flash 2. Send the title + summary + 24 possible topics to Jev 3. Use Jev to classify each paper 4. Visualize everything on https://t.co/hs63SlHxjw The summaries cost $3.99 on @togethercompute. The classifications cost $0.08 on @typesafeai. So for just over $4 of inference, I ended up with a pretty useful way to explore the top AI research papers from the past year. I think this is where things are heading: different models for different parts of the workflow, instead of using one model for everything. I’m running evals on the Jev classifications before replacing the current ones, but the site is already live: https://t.co/hs63SlHxjw

15.9万 views · 1.9k likes · 1.9k saves

Wow, @typesafeai Jev is a game changer for voice-driven UI. Responsiveness really matters for an experience to 'feel' right, and this certainly does the job! Latency here is not optimal given I'm in London and calling out to us-west. But still, super snappy. In this video, Jamcat is using Pipecat to orchestrate a real-time handover between a command agent (Jev) that drives the UI, and voice agent (PhoneLLM) for handling generations and (not in the video) discussing ideas and session setup with the user.

2.6万 views · 341 likes · 337 saves

Claude me integró Jev con Playwright  para buscar usados. Lee unos 26 artículos por minuto y decide qué hacer con cada uno en 406 milisegundos Descartó los que no encajaban con lo que busco, ofertó por los que sí y hasta les mandó un mensaje a los vendedores cuando faltaba algún dato en la publicación. Toda la búsqueda salió USD 0,00085. O sea: con un dólar revisás unas 26.000 publicaciones. Una IA que por fin puede navegar por internet más rápido que nosotros es un game changer

2.2万 views · 356 likes · 356 saves

btw this is what you get if you ask Jev to play Minecraft with GPT-6 Astra. I still can’t believe it https://t.co/ERDnsVWPLZ

4.0万 views · 319 likes · 109 saves

很多人没看懂爆火的JEV是什么？ 所以我做了一个 50 秒的演示视频，带大家把海外刷屏的 JEV 以及它背后的底层架构看透👇（见下方视频） JEV 是前 OpenAI、参与 ChatGPT / RLHF 的 @CompleteSkeptic（Diogo Almeida）创办 @typesafeai 后的第一款模型 他在 OpenAI 时发现了一个致命矛盾： 通过人类偏好微调（RLHF）训练出来的自回归大模型，存在严重的模式崩溃（Mode-dropping）、过度自信与幻觉，且串行逐字生成的延迟极大，根本无法胜任严肃的机器自动化生产 为此，他们花了两年研发了专为自动化设计的 System 1 基础模型 JEV，核心架构创新非常激进： • 单次并行计算（One Parallel Pass）： 如同当年 Transformer 取代 RNN 一样，JEV 彻底抛弃了逐 Token 串行吐字的自回归机制，采用全新的模型架构与采样器，一次性并行击发所有预设选项 • 强化学习校准决策（RL for Calibrated Decisions）： 引入全新的训练算法，输出媒介不再是自然语言文本，而是直接输出经过置信度校准的概率分布 • 代码级可靠（TypeSafe）： 彻底根除幻觉，输出自一致、确定性高，可作为强类型对象直接被程序调用 • 100 倍提速与极端降本： 延迟降至 0.1 秒以内（快到能直接插入 60 FPS 游戏循环）；输入每 10 亿 Token 仅 42 美元，输出 Token 彻底免费 视频里我把传统自回归的 8.5 秒延迟瓶颈、JEV 的并行矩阵运算，以及实时控制小游戏实测，全拉通做了可视化, demo参考 @MatijaSosic 聊天需要慢推理，自动化需要快直觉，Building Prod, not God 确实不是一句空话🫡

4.5万 views · 70 likes · 92 saves

Will Jev @typesafeai turn $10.000 into profit? Let's find out! Running decisions for paper trades via https://t.co/28WVX50sv3 https://t.co/2XglDmGTNM

10.2万 views · 447 likes · 412 saves

I think the implications of jev on self-driving could be huge https://t.co/advoRDSOz7

32.4万 views · 1.5k likes · 593 saves

Built a model router with Jev by @typesafeai. Jev decides what model fits your request best and the request is sent to that model. https://t.co/HHlmjOE66u

10.9万 views · 1.9k likes · 1.1k saves

🚨 I gave the Trump vs Kamala debate a live BS meter using Jev every sentence, both candidates, 5 yes/no questions each 1,191 Jev calls / 1.18M tokens / 415 ms median total cost : $0.0497 same questions for both, clips picked by one fixed rule, not a fact-check https://t.co/m8u4ALl6Me

16.6万 views · 1.2k likes · 489 saves

For those who still didn't understand what Jev from @typesafeai can do, here is the example. Jev can read dozens of structured trading signals and then show intelligent decisions. It's time to make your system actually alive. What a time! https://t.co/r9cfanXcNI

11.6万 views · 966 likes · 1.0k saves

🔥 卧槽了，jev 简直太适合 codex 用户了，强烈建议 codex 用户去尝试下 jev 因为 jev 非常适合需要快速进行复杂决策和评估的场景，所以和 codex 的 computer use 功能简直是绝配。 直接把 jev 用作判官，指导 codex 快速操作你的电脑，既准确又节省 token。 要使用的话需要去官网申请👇，如何申请我已经录制好视频了，跟着操作即可

2.1万 views · 64 likes · 116 saves

I built a browser extension with Jev @typesafeai that can hide/collapse posts on X based on natural language. It's so fast that it's not noticeable and insanely cheap...this must be the future of "ad blockers" and content firewalls. https://t.co/EPIJ4rzeIp

6.7万 views · 1.1k likes · 558 saves

Behind the scenes of our launch video… chief of staff @justKDeng keeping our CEO @CompleteSkeptic on track https://t.co/81HHtsj4eK

4.8万 views · 358 likes · 42 saves

love jev, but upset it - isn't open source? - it lack vision capability? We fixed all of that, introducing https://t.co/DlNeKo2k5e A fully open source library which takes any HF model and Jev-ify it, with an API endpoint Now on github, and live in production at @FeatherlessAI https://t.co/dFKBbR411y

4.7k views · 60 likes · 90 saves

so I gave jev full control of the falcon-9 class rocket full launch to landing no autopilot, no scripted trajectories, no safety veto in the code every decision was made by jev ( you can see decisions at the bottom of the video) and yeah it was able to land the booster and all this with just 245 calls and and $0.04 in API calls for sim I used mujoco and the configurations were 42m booster, 425 tonnes, 9 engines, real atmosphere, mach dependent drag, wind and propellent depletion the vehicle was going ubstable every .8 secs, and jev's decesion making takes place at ~1Hz and output in ~0.4 sec so I gave jev these controls > launch or hold >engine count for every burn( it chose 5 for landing by itself !) > when to light that burn >ascent cutoff, coast, entry, legs, touchdown so after giving all these controls to jev the experiments begun and after 12 runs and tweaking multiple things it managed to land by itself I will write a full article about it and what I did with experiments and whet I tweaks I made but for some final thoughts people are saying it's just a classifier yes, but it's a very good and effective classifier, and its speed and decision making, open up a whole new work for real life tasks and implementationa I will also open source the whole repo for this after cleaning up the clutter in my code so this is the short demo I hope you like it

7.5万 views · 109 likes · 49 saves

i built computer use using @typesafeai ! it is 155x cheaper than opus 5, ~20x faster, and generalizes across OS's more on how it works in the vid & thread below: https://t.co/K0ZaRRnEX5

26.9万 views · 2.3k likes · 2.0k saves

Testing apps with @typesafeai jev & @opencode will be soooo fast small demo: https://t.co/WLSkpe9OZn

50.9万 views · 1.2k likes · 755 saves

Ask Jev anything. Give it a try at https://t.co/P9QMfdDc0r It won't answer. It will judge. Let's see if we can get to 1 million questions. @typesafeai 🤝 @convex work great together. @hmartenjoyer @CompleteSkeptic @justKDeng @mikeysee https://t.co/3efLDDGKqN

9.2万 views · 510 likes · 446 saves

Takie modele jak JEV (System-1) zmienią moim zdaniem bardzo dużo. Obecne modele LLM symulują System-2. Dla mnie implementacja jev to jedna z najświeższych koncepcji ostatnich miesięcy. Niby nic szczególnego (dynamiczny klasyfikator) a dziwi, że tak długo trzeba było czekać na taki model. Uniwersalny klasyfikator, inteligentny "if". W swoich poprzednich wpisach objaśniłem jak to działa. Mam dostęp do API @typesafeai @CompleteSkeptic więc zrobiłem self-play w moim pac-man. To "zabawkowy" przykład ponieważ schemat zapytań jest stały ale ... w wielu systemach np. agentowych schemat jak i stan zmienia się w czasie (w pac-man tylko stan się zmienia). W bardziej zaawansowanych, złożonych systemach struktura stanu, jak i wartości atrybutów stanu są zmienne w czasie, kontrakt klasyfikatora (liczba klas jak i ich semantyka) również. I tutaj mamy system-1 jev - podejmuje decyzje. Oczywiście można to zrobić na LLM ale będzie wolniejsze, droższe, sekwencyjne (wzajemnie zależne - w kontekście widać odpowiedzi). Można wykorzystać dyfuzyję z blokiem odpowiedzi wypełnionym nie na szum a na szablon odpowiedzi (będzie szybko) - generować tylko proby. Możliwości są chodzi natomiast o samą koncepcję Systemu-1. Widzę szerokie zastosowanie - inteligentne aplikacje, systemy agentowe, w klientach do kodowania -> obniżenie kosztów, zwiększenie szybkości działania, jakości rozwiązania itd.

1.1万 views · 95 likes · 62 saves

This is cool. I used Jev to improve my custom memory system. Results: → 94% fewer tokens used → 2–3× faster memory retrieval This was just a quick test, but I believe it opens up so many new opportunities. https://t.co/Xppb20drpY

3.1万 views · 468 likes · 540 saves

Function calling without waiting for Enter. Answers only when needed. @typesafeai Jev understands intent and arguments as you type, then executes ahead of time once everything is in place. https://t.co/n4m9lGGf8S

3.5万 views · 313 likes · 282 saves

You can use Jev to assemble interfaces on the fly. Here’s an example I hacked together with Astra. In this demo, Jev turns natural language requests into UI plans in roughly 1 second or less. It’s 6x faster than Gemini 3.8 Flash and GPT 5.6 Luna. Cheaper, too. https://t.co/SQwcvB3HKh

1.3万 views · 216 likes · 241 saves

thanks to @typesafeai jev I no longer have fill out all of those fields on prompt boxes. It picks the agent / model / computer / folder for me. - For a major rewrite it uses Fable + Claude Code. - Changes to an ios app run on one of my macs https://t.co/3yiKL5EO2w

4.6k views · 97 likes · 55 saves

Silicon Valley predicted Jev a decade ago. https://t.co/ZfSjPYWsTn

32.2万 views · 4.0k likes · 486 saves

UN TÍO ACABA DE LEERSE LOS 1.891 ANUNCIOS DE SU COMPETENCIA POR 12 CÉNTIMOS Bueno, leérselos no. Se lo pidió a JEV, el modelo nuevo de TypeSafe → 1.891 anuncios → 19 segundos → 0,12 $ Y no es un resumen. Es cada anuncio etiquetado por dónde ataca en el funnel y con qué estilo, más la radiografía completa de la cuenta: en qué invierte, qué repite y dónde tiene huecos Saber exactamente cómo vende tu competencia costaba una agencia entera. Ahora cuesta menos que un café Llega pronto al MCP de maxfusion

6.5k views · 77 likes · 31 saves

had to try something with jev so i tried what you'd expect from me: graph extraction - score every word with semantic significance (1-5) - also tag relevant words with ID - unique list of high scoring words - graph of high scoring words https://t.co/oJtI6py84H

2.0万 views · 229 likes · 237 saves

jev is INSANE. we gave it a library of 11 million+ videos on tiktok & ig and asked it to find viral "bloom nutrition" content. in 20 seconds, it gave us over 384 videos, breaking down the hook, format & angle that made each one go viral. (only costed $0.09 in token usage) jev also finds, watches, analyzes and breaks down viral videos and creators, then writes briefs and scripts for you, based on proven data. currently available in @virlomain + MCP. comment "JEV" for access.

6.6k views · 64 likes · 127 saves

📣 Speaker announcement: @CompleteSkeptic CEO, @typesafeai at Cloudflare Connect Register today: https://t.co/YCwgnPlLDj https://t.co/TwT5kMw5bT

1.1万 views · 71 likes · 2 saves

Jev is actually insane. We benchmarked it, and the results completely change the game for us: > 500 real-time agents > running in parallel > in a 3D environment The preliminary results are crazy: > 500ms average latency > 35 API calls/s > all with a naive implementation We did 0 optimizations! The bottleneck is not Intelligence latency anymore. It's the first System 1 LLM. Crazy times.

5.8万 views · 647 likes · 353 saves

Jev made our Slack agent 2x faster ⚡️ Our agent can be quite slow because it needs to read skills and figure out which tools to call. We used @typesafeai's new model to speed this up by first passing it the prompt and classifying the best skill, tool and params to use before handing it to the agent

4.9k views · 60 likes · 42 saves

Played around with @typesafeai Jev today, mostly to understand what it does for tool calling. Instead of an LLM deciding what to do, I substituted that part with Jev. My learning is that we will be able to make the agent act before the user finishes the sentence. So far, we've used different LLM combinations (non-thinking + thinking ) plus state machine setups to get the right experience for the end user. TLDR is to understand more about how voice agents can run decisions on partials instead of waiting for the end of the turn, with lower costs.

4.7k views · 77 likes · 38 saves

Jev enables instant AI suggestions 👏 In this app, 20+ intelligent checks occur every time you edit a cell, immediately providing warnings. No LLMs necessary! https://t.co/brVT7jE6SO

2.9万 views · 200 likes · 183 saves

Jev can draw pictures by predicting every pixel in parallel https://t.co/IOwwmr07cm

30.2万 views · 1.5k likes · 393 saves

Jev solved local harness/model routing I use a combination of Claude Code, Codex and Opencode as my local agentic stack and routing to other harnesses was always enforced in the system prompt/rules With a deterministic hook that Claude Code can decide before delegation, Jev helps to route to the right harness/model based on the task, and it's pretty accurate based on the intensity/intelligence of the task > Mechanical tasks get routed to Haiku > Intelligent ones to Opus sub-agents > Long-running implementation work to external harness

12.5万 views · 814 likes · 877 saves

They say you’re six connections away from anyone on Earth. So I tried Jev from @typesafeai and built this: enter your X handle and someone you’d love to reach. It finds a chain of real follow connections that could get your message there, under 2 seconds! Apparently, Luke Skywalker is just two people away from me. Who’s your impossible person? I can drop the link below if you want to try. @MarkHamill

4.6万 views · 94 likes · 91 saves

We found the perfect use case for Jev: auto-routing GenAI models. The idea: Jev picks the most cost-effective models for your task, balancing speed and quality. All generations made with Higgsfield API. https://t.co/jZWedfbAMN

5.2k views · 36 likes · 16 saves

By now you've probably seen Jev on your timeline. Yes, it's an insanely fast & cheap model. But how does it compare against GPT-6 Astra, Fable 5.1, and Deepseek v4.1 Flash? In this video I compare it against top LLMs on 4 distinct tasks. Did it win? Or did it lose? Let's find out. Total Jev bill for everything in this video: $0.49. Chapters 0:00 Intro 1:07 What Jev actually is 2:57 System 1 vs System 2 thinking 4:36 Jev plays Doom 7:14 How Jev works 10:33 Trading Bitcoin for a year 14:26 1,500 emails at 200 ms each 17:58 857 support tickets vs Intercom Fin 20:06 When you should actually use it

1.8万 views · 97 likes · 78 saves

I build a SQL extension that turns plain english into a WHERE clause with jev @typesafeai WHERE jev(people, 'could work from home') Every row gets judged individually, no index and no embeddings needed. Try it out (don't burn my wallet pls) https://t.co/4nYZVH9iXs

2.0万 views · 106 likes · 66 saves

got @typesafeai’s new model Jev to play Super Mario Bros. fast inference + structured outputs makes it surprisingly good for real time use cases. I'm excited to see what can be done with these new models! https://t.co/wXXW2SsSXG

56.9万 views · 2.8k likes · 1.0k saves

Jev vs GLM 5.3 at chess! Results: ◾ GLM 5.3 won by checkmate in 29 moves ◾ Jev: ~0.3s and <$0.0001 per move ◾ GLM 5.3: ~5.8s and ~$0.008 per move ◾ The whole game cost 24 cents My main takeaway is that it's often useful to use each one to their strengths: ◾ Fast, well-defined classification → specialized models like Jev ◾ Classifications that need reasoning or lookahead → LLMs like GLM 5.3 ◾ Real classification pipelines → hybrid. Jev handles the easy calls, an open model handles the hard ones. The future is multi-model!

4.7k views · 46 likes · 16 saves

The speed and cost is going to enable use cases that haven't been possible before. Here's a quick demo of live decisions of what slide to pull up as I'm speaking, powered by @typesafeai Jev. Also, sorry for the sniffling. I have a bit of a cold. https://t.co/124B5YEjmF

1.5万 views · 99 likes · 90 saves

Another useful use-case with Jev from @typesafeai - AI text classification. Very close to @pangram, 6000x cheaper, 5-10x faster Tested here on a @paulg essay! https://t.co/KijjMknTaC

5.0万 views · 124 likes · 121 saves

Jev is a new model from the co-inventor of ChatGPT that can use computer tools in realtime right now it's playing Minecraft on its own computer in the cloud on Orgo this is what it looks like with just one Jev, but imagine: what about a thousand? or what about a billion of Jevs? each working together on their own computers in the cloud, discovering new science, running experiments, and designing new drugs 24/7. this is going to be bigger than ChatGPT

1.3万 views · 147 likes · 96 saves

Jev by @typesafeai works quite well for drone applications. Made this in 15 minutes and only cost 10 cents. Repo: https://t.co/b2ytFvehTV https://t.co/rC36lr4iqK

13.3万 views · 351 likes · 296 saves

完全睡不着了....这个视频让你完全看懂JEV的恐怖能力 我实际演示了一下，用 JEV 和 DeepSeek 做同一个任务：从当天的海量实时新闻流里，为 15 个品牌快速匹配有没有适合借势的公关热点，并打上结构化意图分类标签 结果是：28 秒内，Jev 狂刷完了整整 428 条数据，并完成了全部的目标分解与分类；而同一时间跑在同样任务上的 DeepSeek V4.1-flash，才刚刚完成 6/428 条 速度被拉开了几十倍，答案甚至在你读完第一行字之前就已经返回了.... 为什么会产生这么恐怖的差距？底层逻辑是： • DeepSeek 这种通用模型本质上是在“逐字写文章”： 哪怕你只让它做个最简单的“是/否”判断，它在后台也必须一个词一个词往外推，硬走一遍漫长的生成流程，延迟按秒起步 • Jev 官方定位是“System 1（快决策）”模型，从根上就根本不会写字： 它彻底抛弃了逐字吐词，底层直接一次性并行出结果。不输出半句废话，70 到 200 毫秒内直接甩出代码需要的最终答案（对/错、打分、分类） demo参考@elvissun Github开源：https://t.co/sdyWJhdvfO 建议大家试试JEV，越用越觉得牛逼.....

4.3k views · 15 likes · 44 saves

I got early access to @typesafeai's new Jev model and built a demo of it playing tetris The generation speed is so fast that it's pushing blocks down This feels like another "this changes everything" moment https://t.co/ND9rVcsPiD

7.6万 views · 966 likes · 274 saves

arkadaşlar jev'e early access aldım ve test etmeye başladım ve gerçekten çok impressed olduğumu söylemem lazım sadece şu kaydettiğim demoyu bi izleyin lütfen. kesinlikle hızlandırılmış bir video değil. 4 tane jev karşılıklı 101 oynuyor ve oyunun hızına bir bakın. beni takip edenler bilir. yıllardır her yeni model çıktığında bizim meşhur 101 okey oyununu oynatırım. benim kişisel benchmarklarımdan biridir ai modellere 101 oynatmak. kimi cahiller ehi ehi ehi adam modele 101 oynatıyor dese de asıl benchmark budur. bizim eve bundan başka benchmark girmez. çünkü 101 göründüğünden çok daha zor bi testtir modelin önce oyunun kendine özgü kurallarını anlaması, sonra oyun boyunca bunlara eksiksiz uyması gerekiyor. ama bu da yetmiyor. elindeki taşları, masa'nın durumunu yani son state'ini ve rakiplerin hamlelerini değerlendirip doğru kararı zamanında vermesi gerekiyor. yani masanın anlık state'ini de her modelin bilmesi ve hızlı karar alması gerekiyor özellikle bu son kısım bugüne kadar modellerin en çok zorlandığı yerdi openai, anthropic vs. modelleri iyi karar verebilse bile bazen 30 saniye veya daha fazla düşünüyor. gerçek zamanlı oyunda iyi karar vermenin tek başına hiç bir anlamı yok çünkü doğru kararı verdin ama sıranı kaçırdın ise pek bi işe yaramıyor. dolayısıyla hız + doğru karar en önemli metrik burada. jev duyurulur duyurulmaz early access'e başvurmamın sebebi de buydu. direkt aklıma 101 benchmarkım geldi jev aslında konsept olarak çok da yabancı bi şey değil. kabaca generalized zero-shot classifier diyebiliriz. uzun uzun metin üretmek yerine kendisine verilen state üzerinde tanımlı seçenekler arasında karar veriyor ama güzel olan bunu ne kadar hızlı yaptığı benim 3-4 yıldır farklı modellerle denediğim ama bir türlü tam verim alamadığım ai'ın 101 oynaması problemini jev çözmüş gibi görünüyor kararları milisaniyeler içinde veriyor ve oyun boyunca oldukça isabetli oynuyor bence asıl önemli tarafı da 101 değil zaten 101 bunun küçük bir demosu computer use, browser use, hız gerektiren oyunlar vs gibi alanlarda yani modelin sürekli ne yapmam lazım diye düşündüğü her senaryoda jev hızır gibi yetişiyor. milyonlarca küçük karar için çok çok ideal gözüküyor milyonlarca mini karar için 30 saniye düşünen dev bir generative model gerçekten işlevsiz. onun yerine 100 milisaniyede yeterince iyi karar veren bir model daha mantıklı jev'in yaptığı da bu bence müthiş bir ihtiyacı karşılıyor ve karşılayacak. hatırlayın 4g gibi hızlı internet altyapıları yokken tiktok vs gibi realtime streamingler mümkün değildi. tiktok'un ortaya çıkabilmesi için 4g ya da hızlı internet gerekliydi. şimdi jev = 4g. hız konusunu çözüyor. dolayısıyla jev'in tiktoku yani jev sayesinde ne gibi yenilikler ortaya çıkacak acaba görürüz önümüzdeki dönem. eğer anlık olarak milyonlarca mini karar alınması gereken bir konseptiniz varsa çok ideal bir şey jev. openai ve anthropic gibi şirketlerin de benzer şekilde generation ile decision'ı zamanla birbirinden daha fazla ayıracağını düşünüyorum

5.5万 views · 169 likes · 178 saves

Controlling Figma with voice using Jev from @typesafeai Total cost from first line of code to recording this video: $0.01 🤯 https://t.co/HTGcrBYpTb

6.0k views · 70 likes · 52 saves

typesafe's jev is fun! live demo you can play with: https://t.co/ctnPF0w70r https://t.co/i6kMGkkWOV

10.8万 views · 584 likes · 458 saves

Jev solved delegation https://t.co/1TznArCaSY

4.3万 views · 441 likes · 341 saves

Jev + Higgsfield = solved auto-routing for genAI models. In this demo, @typesafeai’s Jev evaluates prompt and picks the most fit models for video and image generations on Higgsfield API. https://t.co/MyTRdqlZ3o

3.8k views · 20 likes · 10 saves

这两天关于 Jev 这个模型的热度非常之高，甚至可以说是近半年我重点关注了英推以来，看到的原创内容密度最高的一次。 正好我过去一年一直在跑一个小项目，其中有一个环节就是要对每天出版的《人民日报》上的新闻做一个分类，判断每一篇文章中是否包含有湖北相关的元素。 一年跑下来，积累了接近 2 万 4 千条数据，其中判定包含的有 1800 条左右。在这些数据的基础上，我随机筛选了一个测试数据集，其中包含了 500 条判定包含的内容，以及 500 条判定不包含的内容。 之前我使用的模型一直是 Gemini 的 Flash Lite 系列，也算是经过测试后成本和速度双重考虑下的最佳选择。 今天 OpenRouter 上线了这个 Jev 之后，我用同样的提示词做了一个小 Demo 做了一下对比测试，下面这个视频就是同时并发 16 个进程的实录效果。 不得不说，测试出来的结果超出预期很让人惊喜啊。 首先，速度是真快！相对于 Gemini Flash Lite 单篇文章 3 秒的平均耗时，Jev 把单篇耗时缩减到了 0.35 秒。这个接近 10 倍的速度提升，真的是实现了质的飞跃，带了很多新场景的想象空间。 其次，质量也真不错！和 Gemini Flash Lite 的分类结果相比对，Jev 大概在 15% 左右的内容上出现了判断不一致的情况。 但经过我的手动审核，大部分都是因为人名而导致判断差异。例如某位此前在湖北省任职过的官员姓名出现在了新闻中时，Flash Lite 因为知识库比较大，会将其判断为包含，而 Jev 在这块上面会弱一点，因此判定为了不包含。在我的实际应用场景中，这种判断差异完全可以接受。 最后，价格也是真便宜啊！看了一下 OpenRouter 的账单，前前后后大概跑了 5 千条数据，30M Token，总共开销也就 1 美元！这个性价比，还要什么自行车！ 总之，我个人觉得 Jev 代表的分类器大模型的确是一个非常有前景的发展方向，甚至可以说是在分类场景下实现了速度/质量/价格的不可能三角，有类似需求的朋友真的可以好好考虑一下。 就分享到这了，我得连夜用它去优化生产项目了。

2.9k views · 27 likes · 25 saves

The real use-case for Jev https://t.co/fnsghSM7oo

1.9万 views · 104 likes · 45 saves

Built an alternative version of @typesafeai but on @cerebras with Qwen 3.8 27b. Similar quality, similar performance, but vastly different cost. TypeSafe was way cheaper, and did beat Qwen on performance. Closest we can get using LLMs I think. Source: https://t.co/H5rUhiN7EY https://t.co/aArV0VZE9E

2.6万 views · 208 likes · 194 saves

i made an llm from first principles with Jev 29 yes/no questions per character: should the next key be a–z, space, comma, or period? highest probability gets append to it, then fed the updated text back in &amp; repeat an autoregressive loop made out of a classifier https://t.co/OdiVS5MmM1

16.5万 views · 879 likes · 275 saves

pub-sub without sub @typesafeai jev does the routing almost instantly demo link below https://t.co/pTQIUtVBrI

5.2k views · 62 likes · 71 saves

jev is an insanely cool product Saw a post about jev and decided to check it out This thing is straight up gold whoever built it is a genius jev is the future Scanned over 700 live ads in 40seconds flat, and it cost me just 4 cents. If you're in marketing, this kind of crazy fast data crunching is hands down the best thing out there. Doing this through Opus 5 would run around 2 to 4.5 million tokens, set you back $15 to $50, and take anywhere from 10 minutes to an hour. That’s 200 to 500 times more expensive than Jev, and way slower.

2.6k views · 45 likes · 21 saves

Jev es una nueva IA creada por Diogo Almeida, uno de los creadores de ChatGPT, que promete ser hasta 200 veces más rápida y 400 veces más barata que los modelos actuales. A diferencia de los LLM, no genera texto palabra por palabra: evalúa opciones, toma decisiones y devuelve probabilidades en segundos, lo que podría cambiar cómo funcionan los agentes, las recomendaciones, los filtros de seguridad y muchas aplicaciones de IA. #InteligenciaArtificial #Jev #ChatGPT #Tecnología #EDteam

1.3k views · 44 likes · 32 saves

Silicon Valley predicted @typesafeai https://t.co/v8xXPluYF3

16.5万 views · 1.8k likes · 387 saves

What if your slides are smart enough that they navigate automatically? 👀 Thank you @typesafeai for providing access to Jev! I had this idea for a few weeks, and Jev felt like the right model for this task. (Sorry for the bad audio) https://t.co/uz8UhGBypc

1.7万 views · 72 likes · 31 saves

Jev Founder, Diogo Almeida (ex-OpenAI): "The next era is not the Claude Code era. Claude Code is still part of the assistance era - Jev is the next step Jev is x200 faster, x400 cheaper and has 0 hallucination. This is what comes after RLHF" In 17 minutes he explains why every model you use has a human-shaped flaw baked into it and why Jev is better worth more than a $500 Claude Code engineering course watch today, then read how to actually boost yourself x100 with Jev in the post below

2.5k views · 27 likes · 24 saves

I got access to Jev! It can route website visitors to the pages they need based on what they need help with! https://t.co/AydblKM9Y0

2.6万 views · 291 likes · 223 saves

Introducing the Papers with Code MCP server! To celebrate its release, I've asked Claude Code to research the architecture of Jev by @typesafeai using the `search_papers` tool. Here's what it came up with: "If Jev is architecturally anything public, it's a large schema-conditioned bidirectional encoder with parallel label-query heads (GLiNER/ML-Decoder shape) trained with a proper-scoring-rule RL objective (RLCR shape), scaled far past the ~150M-param range those papers operate in — "neither small nor an LLM" fits that. The 40–200× speedup is consistent with removing autoregression, not with any exotic mechanism."

1.5万 views · 78 likes · 57 saves

I just built Easy-Jev, a Jev demo anyone can try right now! Change the inputs and watch the classifications change in real time. I did my master's thesis on classification models so it's fun to see a modernized version of a very useful alg Typesafe are correct in that LLMs are just one approach to intelligence, and I think as time goes forward we’re going to see a hybrid of dif models splitting up and allocating correct parts of a problem, using less tokens overall and leaning on LLMs for less of the answer. I’m bullish on what I call 'combination input' approaches: breaking a statement down into a series of classifications and routing each to the right type of model rather than just brute forcing the whole thing with an LLM cc / @CompleteSkeptic @typesafeai Link below, please enjoy

1.1万 views · 87 likes · 73 saves

Acabo de terminar la implementación de @typesafeai + Chromium Headless para que mis agentes puedan navegar por internet a una buena velocidad! En este ejemplo le pido al agente que entre a la página del término "Café" en Wikipedia y navegue por los hipervínculos hasta terminar en la página "Inteligencia artificial" Logra completar la prueba en 20 segundos. Creo que pocos humanos podrían hacerlo en ese tiempo.

2.5万 views · 212 likes · 216 saves

Had jev, claude haiku 4.5, and gemini 3.5 flash-lite play tetris. Same 200 pieces, same legal moves, real time. Jev: 9200 pts, 75 lines, 300ms/move, 0 errors Claude: 8900 pts, 72 lines, 1.52s/move, 0 errors, $0.48 Gemini: 9000 pts, 75 lines, 1.13s/move, 0 errors, $0.02 Jev wins on score and is doing it 4-5x faster per move than either LLM, for basically free. That gap isn't the model being smarter, it is the model doing less work per decision

2.0万 views · 122 likes · 41 saves

Alright @typesafeai cooked Jev is truly the next era of building w LLMs Added this to Scappa tonight https://t.co/YeYnj5pc7l

2.7万 views · 319 likes · 77 saves

昨晚申请的 Typesafe AI 账号，半夜已通过审批！ 幸运之星在此感谢 @typesafeai @CompleteSkeptic 今天抽空写了个 Chrome 插件，通过 Jev 标注 X 平台中帖子内容的意图和概率。 实时判定，并将分类结果以标签的形式展示在时间戳的后面。 分类的类别包括诱导、挑拨、推销、机器生成、说服、娱乐、告知等。 Jev 太疯狂了，几乎秒响应，这在 LLM 上是无法感受到的。 而且我还未细调的情况下，它的分类就已经很精准。 在右侧调出了统计面板，展示大致的使用情况，供大伙参考。 接下来，我会继续将 Jev 整合到更多的产品和工具中。 Jev 是 2026 年度的 AI 之星！

7.1k views · 18 likes · 15 saves

JEV CAN DRIVE! 🤯 I hooked up Jev to the raw controls of a driving simulator, this video shows him driving in realtime. Realtime is the kicker here. Asking a model what to do every 50ms then rendering that into a "realtime" video is one thing, but in this simulation I wanted to give Jev the challenge of actually controlling a moving vehicle in a simulation that will never pause while he thinks.

4.8万 views · 281 likes · 103 saves

jev + kernel browser use demo pretty freaking cool! so fast! try it yourself: https://t.co/CoVFzscaa0 https://t.co/OyXdNA1H0F

3.5万 views · 238 likes · 213 saves

Big AI news... Jev enables this microphone that mutes you when you talk BS @typesafeai https://t.co/VWyJHJ7Oh9

2.7万 views · 176 likes · 57 saves

I've had access to Jev by @typesafeai for about 30 hours now and I can't stop thinking of the million places it can be used 🤯 Meanwhile, checkout this fun snake game I built where 2 Jev's compete with each other to take the trophy 🏆 Try it out here 🔗 https://t.co/axrG1uKIhr It's open source btw The low latency made it all possible to make such a fun project which requires input in real time 🚀

3.3k views · 16 likes · 13 saves

I got access to Jev and immediately plugged it into Claude Code to see if it actually FIXES the biggest pain of agent workflows: context compaction > result: pretty wild Quick context for those who missed it: Every Claude Code session has a context window that fills up fast with tool calls (file reads, grep outputs, logs, etc). When it fills up, the agent runs "compaction" - asks a separate LLM to summarize the chat. this is slow, expensive, and the LLM often drops important context or bloats the summary with garbage: fast-jev-compaction uses Jev instead Link: https://t.co/WD7kpBfDwL Why this works so well: Instead of generating a text summary, Jev scores every tool call in your chat with "how important is this?" probability. > above threshold → keeps it > below → drops it or replaces with a stub > in summary: no text generation, no bloat, no dropped critical context What you actually get: 1/ way faster: Jev outputs probabilities in parallel, not sequential tokens. 2/ way cheaper: 1M tokens of context = ~$0.04 to compact. 3/ deterministic: no "creative" summary that removes half your codebase context. btw, TypeSafe gives $5 free credits at enough for weeks of testing. Link: https://t.co/NzGPArgLvc Honest caveats: Jev only sees the tool CALLS, not their outputs, hard to say if compaction quality is objectively better than default. Jev's context window is small (32K), so 150K+ chats need to be chunked. This isn't a full replacement - it's a smart pre-filter but as a proof of concept. This is exactly the kind of surgical fix that makes agent workflows viable at scale. Now thinking about what else in my stack could get 100x faster with Jev as the decision layer.

2.5k views · 31 likes · 16 saves

I got access to Jev by @typesafeai today morning and I built a cool use case for it Introducing DiffJury - simply paste any public PR link and Jev tells you immediately if it's safe to merge or does it require review ✅ 🔗 Feel free to try it out here - https://t.co/XxcZxIZJDK Imagine Jev being able to tell you if you should merge a PR with grounded context of your codebase. that's what we're building at @graphify 👀 It's fascinating how insanely fast Jev is... the model architecture in itself is quite interesting and this has opened up a plethora of new use cases and I'm sure the internet will pick up on it sooner than anyone'd expect

8.6k views · 79 likes · 61 saves

i used jev from typesafeai to build a model router cli you give in a task and current ai subscriptions you have, the jev tells you which agent/model should pick it up pair this with herdr and it will be a killer combo jev was the missing bit that makes this whole thing possible by being a super fast judge

6.1k views · 65 likes · 50 saves

TYPESAFE RELEASED A MODEL THAT DOESN'T GENERATE TEXT, AND IT BREAKS THE COST FLOOR OF THE AGENT LOOP jev is what they call a system one model. you send it a state and a set of questions, it returns typed answers with probabilities. no tokens, no sampling, no sequential decoding. route → decide → guard → run tool → observe → done? they report 20-200x faster inference and 40-400x lower cost than comparable llms on classification tasks. every question in a request is evaluated in parallel, so adding questions barely moves latency. in a two-turn loop that means eight decisions. the two that need a plan still go to a chat model. the other six - pick a model, screen a tool call, check if the task is done - cost 0.06s combined. the takeaway is simple: stop paying generation prices for choices. most of what an agent decides is a pick, a score or a yes/no.

2.0k views · 29 likes · 17 saves

Jev is already on OpenRouter!! It's only been out a day and TypeSafe still has a waitlist. OpenRouter has it though, and that's how we've been running it. The pricing is insane. $0.042 per million tokens in, and output is free. How to use it: - Sign up on OpenRouter and add some credits - Make an API key - Call the decisions endpoint: /api/alpha/decisions - Set the model to ~typesafe/jev-latest - Send it a "state" (what's going on) and "questions" (what you want decided, plus the options) ChatGPT and Claude write you an answer word by word. Jev doesn't write at all. You give it the options and it sends back a probability for each one, plus how confident it is. What we got: - "My card was charged twice" went to billing, 100%, in about a second - We hooked it up to Tetris and it picked every move in about half a second - A whole game cost less than half a cent A lot of fun stuff can happen here.

1.2k views · 22 likes · 12 saves

Jev is insane for ad research. I built a tool that scores every shot of any video ad on meta in 1.5 seconds. It runs on Maxfusion + @typesafeai I pushed 450+ ads through it in less than 3 minutes. It pulls any advertisers live video ads from Research Lab on Maxfusion. Maxfusion watches the ads and gives a forensic breakdown. Jev then judges each shot and rates the whole ad. It made 50 judgments in 1.52 seconds for $0.00028 and scored this Duolingo ad at 88. Jev is the fastest decision maker I have ever worked with.

3.4k views · 12 likes · 23 saves

Jev opens up so many possibilities for gaming 🤯 🏁 I built Wordshift - the first semantic typing racer game built with Jev from @typesafeai Slower than a jet, bigger than an elephant, faster than a cheetah. The bigger the difference, the farther your car moves. https://t.co/elObRrcVti

5.7k views · 32 likes · 14 saves

I got access to @typesafeai's Jev model and got it to play Pac-man Jev decides which direction Pac-Man should turn at each junction, given the maze state as JSON, and plays in realtime. Demo link → https://t.co/GpfiB6MjHP https://t.co/3bKTb2FsZr

8.6k views · 79 likes · 31 saves

i got early access to jev by @typesafeai and decided to build a little extension that judges my tweets while i type them it runs so fast it can react to each keystroke, and it's basically free, after testing for an hour it cost me half a cent 🤯 https://t.co/HRU6q9kunq

4.4k views · 58 likes · 19 saves

I was wondering if @typesafeai Jev could generate images so here's what a sunflower looks like according to Jev : https://t.co/vT4UiV4SqJ

4.4万 views · 140 likes · 49 saves

Jev (@typesafeai) 使って色々プロダクトで使えそうなの考えてみてる。 dndをしたら自動で仕分け先を教えてくれる体験 最初一回はちょっと遅いが、そのあとはサクサク https://t.co/2Z7K8FA9XI

4.0万 views · 181 likes · 100 saves

Building got faster. Taste didn’t become optional. Framer is where frontier teams like @typesafeai still go when the site has to look like a serious company on day one. https://t.co/cHYyiBhiGW https://t.co/8Vv0Yco3cr https://t.co/N4X1SL6lQ1

7.9k views · 68 likes · 9 saves

@typesafeai playing minecraft, costing only 1 cent (150k tokens) for 2 minutes of play automatically knowing to dodge zombies at nightfall with no prompting https://t.co/kAlHjh2Dfg

12.5万 views · 266 likes · 108 saves

I gave Jev a robot arm. Goal: red cube in the green box. Catch: the cube was out of reach. It grabbed a hook, dragged the cube closer, put the hook down, and finished the job. Tool use was never programmed, the model just decided. Fast enough to run a robot 👀 @CompleteSkeptic @typesafeai @dotpem

4.2k views · 77 likes · 54 saves

I gave @typesafeai the most important binary in @SiliconHBO history. hot dog or not hot dog Try here: https://t.co/m6LmF6svsB Powered by @render workflows https://t.co/SvN7AniSBc

3.1k views · 25 likes · 4 saves

Live @typesafeai Jev demo at @AgentConf https://t.co/y3UpKP2e4z

1.2k views · 20 likes · 3 saves

Just hacked a fun little project - Skill Router (Open source) using Claude and Jev @typesafeai Why? I have ~ 90 Claude Code skills installed and use maybe five. The rest are fine, I just forgot they existed, and Claude won't reach for a skill on its own unless you call it. So I built myself a little personal skill-router: https://t.co/A8HdsngGXf https://t.co/yEmWwAMO1q

1.8k views · 22 likes · 7 saves

Small update for https://t.co/xXR4JLt377 today: 1. minirouter/fusion is live. It classifies each request with Jev by @typesafeai, then routes to a handpicked model matched to the task's type and complexity. 2. Codex now lists your presets and our routers in the model picker. 3. Fixed long prompts not caching automatically in Claude Code. BYOK, sandboxes, compute passes and compute yield coming next.

3.3k views · 67 likes · 2 saves

Built the ultimate syntax highlighter using @typesafeai Jev (i mean why not). It can highlight any language, even one that I made up.

4.1k views · 47 likes · 23 saves

. @typesafeai のJevを使って、Local Whisperの文字起こしを都度判定して、Macbookのローカルコンピュータの操作を自動化して、あれこれするのやってみました！ Chrome のページは Browser Harness、Slack やNotionは Mac の Accessibility で認識して操作してる。 実装はGrok4.6のみ(Codexの制限きてたんで）

2.4k views · 6 likes · 7 saves

Ran a dumb experiment last night: used Jev by @typesafeai (a decision/classifier model, not a text-gen model) to write C++. At every step I gave it the finite set of grammatically valid next tokens and let it just... pick one. No free-form generation, no hallucinated syntax, purely "choose from this legal set." It's not built for this, that's what made it fun. Got a working piece of C++ out of a model that fundamentally does not generate text. Not claiming this is how you should write code, and it's slower and dumber than just using an actual code model. But it's a neat proof that a classifier constrained tightly enough can stumble into looking like a generator. Novelty over utility here, but it's the kind of dumb experiment that tells you more about a model's actual guardrails than the official demos do.

1.2万 views · 49 likes · 18 saves

Jev is insane. We are building egocentric training data for physical AI. Jev QA'd 58,643 action labels in under 3 minutes. 90 cents. A friction cost of the Claude &amp; GPT models. https://t.co/s7n5NpsLya

1.1k views · 15 likes · 17 saves

Jev from @typesafeai plays @BalatroGame. Every hand it sees the cards, jokers, and blind, then picks play or discard in ~200–500ms. After the ante, it shops too. A full run costs less than a cent. https://t.co/PyGvC0dvga

1.5k views · 4 likes · 2 saves

Had some fun taking Jev from @typesafeai for a spin, switching seamlessly between manual and AI-driven gameplay. It’s clocking 2–3 calls/sec at 60 FPS, with ~150ms avg latency. The most interesting part wasn’t the speed, though. It was distilling the game into the variables, actions, and decision criteria the AI needs to play well: Does it risk going for the powerup, or protect the ball? When is the reward worth the risk? Fun times ahead!

646 views · 14 likes · 4 saves

Got early access to @typesafeai Jev and spent a few hours building on it. The problem I keep hitting: agent traces tell you which tool ran, how long it took, what it returned but never whether the agent is actually getting anywhere. An agent editing, testing and reverting the same file six times looks perfectly healthy in the logs. So I made JevScope. It watches an AI agent work and asks Jev what each step actually means is this aligned with the task, is it progress, is it repeating itself, is it stuck. Every step, plotted over time. Here's a coding agent getting stuck on a race condition. Watch the yellow line. No log line says "I'm stuck." The curve does. Detailed demo, code and writeup coming soon.

1.2k views · 8 likes · 1 saves

o modelo jev (paradoxo de jevons) com RLCD (reinforcement learning for calibrated decisions) da typesafe ai https://t.co/aGwtQkbJyj

370 views · 11 likes · 7 saves

it's f*cking insane Jev sat between GPT and me, killing every draft that broke my rules before i saw it. good setup. then Jev did not answer. the agent decided silence was safer and stopped sending me anything at all. took a second agent to unstick it. your checker needs a branch for the moment it goes quiet: > let it through tagged unverified and keep moving > hand the risky span to the big model > park it in a queue and retry in a minute > wake a human when the action is irreversible 70-500 ms one Jev call, by TypeSafe's own number. a half-second checker still takes down the whole pipeline when it has no default.

215 views · 18 likes · 13 saves

One of the popular requests since launching @thallyio is a demo of how it works. So here it is! Thally can create a new docs repository on your GitHub account, connected to the products it documents, and other secondary surfaces that needs to be kept in sync as the product evolves. Now exploring how we could benefit from @typesafeai to make the workflow even more efficient

693 views · 19 likes · 1 saves

i have created moderation guard based on typesafe's jev its fast and accurate live demo you can: https://t.co/FquTJtWCzq https://t.co/PyNNZGu0iW

2.5k views · 21 likes · 15 saves

@typesafeai jev playing chrome dino! the model is incredibly fast! - avg response time: ~280ms - spent 116M tokens - cost? just $4.65. - the input here is a snapshot of the chrome runner - jev decides whether to jump, run or duck with a confidence score repo link if you want to try it out! 👇🏻

619 views · 34 likes · 2 saves

Spent the evening making sense of this, so here's Jev in 34 seconds. Super excited to build some incredible things with @typesafeai what a time to be alive... https://t.co/Zu44BEhsyh

3.5k views · 52 likes · 25 saves

9 Use Cases I thought of for Jev Hope you find it useful

729 views · 5 likes · 7 saves

Jev by @typesafeai turns your  Mac into an AI agent... what this means for You ? - Jev, a new model from a ChatGPT co-inventor, can operate computer tools in real time. - It is 200x faster for Computer use and 400x cheaper in addition to the optimizations. - Yappy from @MitosisLabs is an AI that lives on your mac and do several things in parallel. Filling out Job applications, doing your Taxes and other documents, filling out forms, and so on... - With Jev, all these things happen at the blink of your eye! Below is a video demonstrating this IRL. Here are the benchmarks (Yappy + Jev V/S our competitor): - End-to-end time: 1m 54s vs 13m 14s - LLM Calls: 22 vs 104 - Input tokens consumed: 0.39M vs 11.24M - Estimated cost of the run: ~$0.24 vs ~$4.24 Link to download in comments.

324 views · 8 likes · 3 saves

I used Jev to make an audience of 100 personalities to yap to. Each blob has its own personality and makes its own Jev call every time you talk to decide if it’s bored of you yet. Each round costs &lt;$0.01 in credits. Really cool model from @typesafeai ! https://t.co/nGlKuxxfAA https://t.co/YgRdcS5pfs

1.4k views · 13 likes · 2 saves

Testing @typesafeai Testing TypeSafe as the decision layer for an AI game companion. It reads the player’s message and game context, then returns an action, destination, and confidence. Game code makes the character act. Dialogue is prewritten; TypeSafe decides what happens next.

4.2k views · 45 likes · 12 saves

Jev unlocked a new way to stop AI agents from spinning. Agents do not always repeat the same action. Sometimes they try different things based on the same wrong assumption. I built a semantic circuit breaker around that. In one stuck run: 12 model calls -> 4 19.9K tokens -> 5.1K https://t.co/fR44bfNDrA

667 views · 6 likes · 4 saves

Another brilliant launch for developers: and its 20-200x faster than LLMs because it skips token-by-token generation entirely. TypeSafe AI just launched Jev, > 20-200x faster >40-400x cheaper (w/ output tokens free) > Frontier composable intelligence optimized for decisions So Jev is an AI model built to make software decisions instead of writing text for people. A normal LLM answers by generating tokens sequentially, so software often has to request structured output, parse it, validate it, and decide what happens next. Jev removes that translation layer: give it some data and a predefined question, and it returns a typed choice or score with probabilities and confidence. For example, a support app can ask whether a ticket is urgent, whether it violates policy, and which queue should receive it, then act on those answers directly. In code, Jev behaves like a smart if-statement: ordinary software controls the workflow while the model handles fuzzy judgments that rigid rules struggle with. Jev evaluates multiple structured questions in parallel rather than writing an answer token by token. TypeSafe reports 70-500ms responses and 40-200x faster performance than comparable LLMs

1.7万 views · 144 likes · 63 saves

Built a Grammarly-like Mac app with Jev. Jev is TypeSafe’s first “System One Model”, built for fast, structured decisions rather than generating text like an LLM. I wanted to see how fast that feels in a real app. https://t.co/XIFYnHXO7Y

3.6k views · 7 likes · 5 saves

another demo for jev for ~instant variable generation https://t.co/ddjvvHXpai

380 views · 7 likes · 1 saves

So I ranked every AI model till date in a tier list from S+ to F using Jev from @typesafeai into an ULTIMATE AI TIER LIST for General Intelligence, giving a lot of weight to both intelligence and performance Agentic Tier ranking coming up next! https://t.co/t54lXmbUij

207 views · 10 likes · 3 saves

ChatGPT's co-inventor Diogo Almeida says his new model Jev is 193x faster than AI. Jev by TypeSafe never writes text. Context in, typed decision out. YES or NO instead of an essay. You are paying a chat model to think for 9 seconds about a yes or no.

2.9k views · 28 likes · 20 saves

@typesafeai jev is 5x faster and 1000x cheaper than astra at playing 2048! though it (clearly) does not excel at reasoning compared to state of the art models like astra and fable what should we try next? https://t.co/24Re9v7TRT

3.4k views · 29 likes · 8 saves

Zero-shot robot tasks with the @typesafeai api Give it a goal in plain English. Jev chains a set of hard-coded primitives. The arm does the thing. No training required. More in the thread below 👇 https://t.co/zp8dSy40SW

1.8k views · 7 likes · 5 saves

man who co-invented ChatGPT just shipped a model that does in 0.114 sec what GPT Astra does in 8.5. i wired Jev into the routing layer of a build i was running on Astra: same task finished 37 MIN SOONER, zero malformed responses to retry. you hand it a situation and a list of typed questions. it answers all of them at once, in parallel, with a confidence number attached to each. no chain of thought, no tokens streaming out one by one. you stop parsing a JSON blob back into your code and praying it validates, because the model physically cannot answer outside the options you defined. the numbers TypeSafe publishes for it: > $0.042 per million input tokens, and output tokens are free > same accuracy as Sonnet 5 on their eval, 67.8% for both, at 294x less cost and 195x the speed > 0.114 seconds against 8.566 for GPT-5.6 Terra on the same demo > their Doom demo has it making around 10 decisions per second $40M seed led by DCVC, and the name is a nod to Jevons, the economist who showed that cheaper coal meant burning more of it, not less. this runs underneath the agent, on the calls that happen ten times a second and never get read by a person. waitlist for the API, but the browser playground is open now. links below.

2.0k views · 37 likes · 23 saves

I built a trading agent with the all new @typesafeai Jev. You can try it here : https://t.co/e7bLjnCfTj. Just put your own API key and start playing or clone it from https://t.co/7nJeuB2H8t and test it locally. To my surprise the agent was able to take a trade of 2.405 $ETH and took a profit of roughly 0.23%. It is very low but testing a new model in town is genuinely crazy. @CompleteSkeptic did you guys tested your model like this ? Let's #connect if you're also into fintech and actively researching.

648 views · 12 likes · 3 saves

Got @typesafeai’s Jev playing Minecraft Skyblock. One agent, reading the game state and choosing its next move through typed decisions. https://t.co/RXmybNhVtB

758 views · 14 likes · 2 saves

tested Jev as a real-time robotics policy in MuJoCo it struggled at first, so i split each update into two calls: decide what to do next, then decide how to move the arm and gripper Jev doesn’t accept images, it gets simplified geometry and contacts as text here https://t.co/iKK7jcrrpO

150 views · 3 likes · 2 saves

On the Jev waitlist? You can use not Jev (but close enough) right here: https://t.co/S0EQnHkSGL https://t.co/qLtRs0fHyD browser demo https://t.co/BUFN88boT4

3.1k views · 10 likes · 3 saves

we spent 3 years learning how to write prompts Jev from @typesafeai breaks big prompts into smaller decisions you can build around instead of one giant prompt hoping the model gets everything right, you ask smaller questions: - is this urgent? - is this billing? - does a human need this? then combine the answers in your own code less prompting, more building

3.6k views · 59 likes · 17 saves

Meet Jev (Typesafe's First "System One" model) who knew the next step in AI would be making LLMs obsolete... An AI fast enough for real time applications (including playing Doom in real time) where do i sign? https://t.co/NNcxfwbvfu

152 views · 8 likes · 1 saves

built a chrome extension that lets me write my own X algorithm. it hides engagement bait, empty hooks and ads before I see them. the rules are plain English. Jev by @typesafeai checks every post in half a second. anyone want to try it? https://t.co/ybO4pSm1pK

170 views · 2 likes · 1 saves

#jev it turned out it's good in snake too! fast too! Survived 321 moves! #demo #day_2 #type_safe https://t.co/OFcdy6bsLH

539 views · 0 likes · 0 saves

Built an SO-101 pick and place simulation using GPT-6 Astra and @typesafeai Jev. Astra reads overhead and wrist camera images and produces structured observations and Jev returns typed decisions with probability scores. A separate local controller carries out the movements and checks that the arm stays within its limits and is ready for each step. The models never directly control the motors. A fixed pick and place sequence probably doesnt need this much decision making. I m using it as a testbed for situations where observations are uncertain and the system needs to decide whether to act wait or stop.

453 views · 8 likes · 1 saves

The Jev Doom demo by @CompleteSkeptic is really cool, so while I’m waiting for access I tried the same thing with Gemma 4 E2B from @GoogleGemma. Mine takes 640×480 frames directly, with no intermediate text description of the scene. It shares the image prefill across batched action queries, then reads the controls directly from logits. It sort of works.

521 views · 10 likes · 1 saves

Is Jarvis finally here? TypeSafe just launched Jev, an API for structured decisions that can route models, rank search results and check other AI systems WITHOUT a chatbot in the loop. Imagine a Magic 8-Ball that actually works: a typed answer, with odds and a confidence score. This means they can: • Model routing: send each request to the right mode • LLM output checks: flag weak answers for review before users see them • Support workflows: sort tickets and decide which to automate or escalate And just realized TypeSafe reports ~150ms latency, ~100x lower cost & faster responses than LLMs. Early API access waitlist is open below.

7.0k views · 3 likes · 4 saves

Told my hermes agent to test the @typesafeai Jev model on an iOS app I’m currently developing for OSS Hermes recorded the screen and sent it to me on telegram I’ll continue do some more testing on this model This was the output https://t.co/d9KuwDdfCm

289 views · 2 likes · 0 saves

ok so you can also make Jev auto-decline cookie banners and pop ups now here's a demo of it working on different websites (gymshark, fashionnova, nypost) https://t.co/QZbYn701Vb

95 views · 3 likes · 0 saves

Finally I got access to @typesafeai (thank you guys!) Jev + emulator lookahead cleared Super Mario Bros. 1-1 only 0.04$. In this setup, Jev sees no pixels: I parsed RAM into structured state, simulated controller macros, filtered predicted deaths, and let Jev choose. Recorded the run; resumed from the same state after an API outage. The tradeoff: RAM gives precise, auditable observations, but my adapter is game-specific. Bypassing vision makes this experiment easier; it doesn’t solve perception for general game agents. The potential is Jev as a tactical decision layer over structured observations and predicted outcomes. Caveat: 1-1 is easy for RAM + search. Next: Jev vs a cheap heuristic on identical simulations. Does the model actually earn its place?

296 views · 2 likes · 1 saves

Yes, but can Jev do multi-agent swarms? Enterprise solutions that self-heal live productions before clients can even refresh their browsers? (Demo slowed down for humans) https://t.co/Ky7xNqziss

33 views · 1 likes · 0 saves

THE JEV DEMOS SENT ME DOWN A RABBIT HOLE now i have 144 youtube uploads turning into a content map on one screen here’s the concept i built ↓ 144 real videos from 4 AI channels → thumbnails form a wall → titles get tagged by topic and hook → the wall reorganizes into a topic map → three video ideas close out the sequence the part i like most: you can see the source material behind the ideas this is a concept demo with precomputed, title-based labels — not a live Jev run or a speed benchmark next experiment: pick one idea, build it, show the result bookmark this

17 views · 0 likes · 0 saves

Made Jev fight OpenJev in a first person shooter. https://t.co/wSB3psnvwB https://t.co/vwewLu682T

16 views · 0 likes · 0 saves

Make music in real time, just by describing what you want. A demo I build today with JEV and GPT-Live-1 👇 https://t.co/ndWYSybxiM

5 views · 0 likes · 0 saves

Jev can generate a fight scene on your webpage. It asks each block: floor, button, or ignore? Then each hunter asks: chase, wait, flank, or retreat? https://t.co/CR92Rn3aRt

4 views · 0 likes · 0 saves