Together AI Fixes a Hidden Bias Crashing LLM Reinforcement Learning Training
A new additive correction called score centering removes the hidden drift that destabilizes off-policy RL when training and inference engines disagree.
// Intelligence
Today in AI
A focused feed of models, agents, research and open-source releases for people building with AI.
Updated
Vals AI's Claude Opus 5.5 Agents Proved a Faster Shortest-Path Algorithm
Ten Claude Opus 5.5 agents collaborated on a message board for 15 hours to produce C-HD, a Lean-verified shortest-path algorithm that beats published bounds in a specific density regime.
OpenAI Drops GPT-6 Sol and Luna at 50% Less Than GPT-5.6
OpenAI expands its GPT-6 lineup with two cheaper, faster siblings to Astra, cutting API prices in half and pushing prompt caching further.
How to Review Your Vibe Coded App Before Deployment
METR’s API key theft shows why you must check authentication, credential access, and permissions before deployment.
GitHub Integrates Anthropic's Claude Opus 5.5 Into Copilot at 40% Lower Cost
Anthropic's newest flagship model lands in Copilot Pro+, Max, Business, and Enterprise, promising Opus 5 quality with fewer steps, fewer tokens, and 40% lower cost.
Anthropic's Opus 5.5 Tops Cursor's Coding Benchmark at 40% Lower Cost
Anthropic's new flagship lands in Cursor with a 57.8% CursorBench Max score and 40% cheaper per-task cost than Opus 5.
Lovable Adds Opus 5.5 and Builds Apps With 57% Fewer Steps
Lovable swapped in Anthropic's Opus 5.5 across its app-building agent, cutting build steps by up to 48% with no drop in output quality.
Anthropic's Opus 5.5 Cuts Agentic Coding Costs 40% While Running 30% Faster
Anthropic's new flagship model matches Fable 5.1 quality on most work while cutting costs 40%, with faster output and stronger alignment scores.
Runway Launches DIFFUSE to Connect AI Video Creators With Studios
Runway's new DIFFUSE platform turns its Talent Network into a full hiring marketplace, matching AI-fluent creatives with brands, agencies and studios that need them.
White Circle's Halo Open-Source Framework Trains 20B Models 2.8x Faster
White Circle open-sourced Halo, a distributed training framework that keeps models in native HuggingFace format while hitting up to 2.8x TRL throughput.
Microsoft Shows 5 Communicating Agents Matching 33 Independent AI Attempts
A Microsoft Research and UC Berkeley team shows agents sharing a scratchpad beat parallel independent runs, setting new records on ARC-AGI-3, polyomino packing, and MNIST compression.
Moonshot AI Turns Kimi Browser Extension Into a Chat-Driven Web Agent
Moonshot rebrands Kimi WebBridge as the Kimi Browser Extension, adding a sidebar chat, recordable skills, and cross-agent support that runs locally.
Tencent's IWC-Bench Actually Runs AI Web Apps to Catch Broken Buttons
Tencent Hunyuan's new interactive benchmark actually runs AI-generated web apps, uses code coverage to hunt broken paths, and matches human taste 85.3% of the time.
Jev Chat Assistant Drafts WeChat and QQ Replies Using Android Accessibility
An Android accessibility-service app reads WeChat, QQ, and X conversations, then drafts three ranked replies you can paste with one tap.
Amazon Brings Moonshot AI's Kimi K3 to Bedrock at 1.76x OpenRouter Prices
Moonshot's 2.8T-parameter open-weight model lands on AWS with a 1M-token context, native vision, and explicit prompt caching for agent workloads.
Tencent's Hy Image 3.5 Merges Generation and Editing Into one Model
Tencent's Hunyuan team launched a preview of Hy Image 3.5, claiming a 30% win rate boost over 3.0 at $0.024 per image.
Artificial Analysis' Benchmark Reveals Gemini 3.1 Flash TTS Beats Sonic 3.6 on Pronunciation
A new benchmark scores text to speech models on how correctly they pronounce tricky words, and voice preference does not predict accuracy.
StepFun's Step 5 Preview Matches Top Rivals at One-Third the Cost
StepFun's new 600B MoE flagship matches Kimi K3 on the Intelligence Index at roughly one third the cost, with open weights due in October.
xAI's Grok 4.7 Beats Claude at Enterprise Analysis but Doubles Your Bill
xAI's Grok 4.7 jumps 111 Elo on Artificial Analysis's agentic knowledge work benchmark, closing the gap with Anthropic at roughly half the cost per task.
Cognition Adds xAI's Grok 4.7 to Devin but Warns of Scope Creep
Cognition ships xAI's newest model in Devin CLI and Desktop, where it excels at hard backend work but over-scopes on simpler tasks.
Xiaomi's MiMo-V2.6-Pro Tops Open-Weight Rankings With a 1T-Parameter Model
Xiaomi's new 1.02T parameter MoE tops the Artificial Analysis open weights leaderboard while costing pennies per task through aggressive caching and sparse activation.
Meta's Petal Cable Will Move 1 Petabit per Second Across the Atlantic
Meta's Petal cable will deliver 1 petabit per second across the Atlantic using multi-core fiber, doubling capacity over the most advanced systems.
Dots Studio's Dots3-Note Preview Tops Open-Weight AI With 76.8% on ARC-AGI-2
Dots Studio's 280B mixture-of-experts model hits 76.8% on ARC-AGI-2 at eight cents per task, topping the open-weight verified leaderboard.
Inco AI's Splash Runs Qwen3.8-27B Twice as Fast on Apple Silicon
Inco's open-source Splash engine runs Qwen3.8-27B at 74 tok/s on an M5 Pro, hitting 2x oMLX and 3x Ollama on Apple silicon.
No stories match these filters.