A focused feed of models, agents, research and open-source releases for people building with AI.
Anthropic released FlashProve , open-source cryptographic proofs that a response came from a specific LLM. Generates proofs for 400B Llama 4 Maverick at 110 tokens/second on a single H100 GPU. Proofs are ~19 MB and verifiable in seconds on an ordinary 24-core CPU without weights.
Sarvam released Prism , a 105B decision model under Apache 2.0 that outputs probabilities instead of text. Median latency of 44 ms per request on a single NVIDIA B200, 95th percentile at 420 ms. Scores 71.8 average across three decision indices, claimed highest for sub 100 ms models.
humans& released GPU-accelerated Weight Teleportation , lossless RL weight sync faster than intra-cluster RDMA. Sampler pause drops to 219.6 ms; end-to-end sync under 4 s on 445 GB NVFP4 GLM-5.3. Snappy plus Zstd double compression shrinks per-step delta payload to 2.65 GB, 0.60% of model weights.
Edge0 released Audio8-TTS-Realistic-Air , a 0.6B TTS model with zero-shot voice cloning, Apache 2.0 licensed. Supports 23 languages including Arabic, Hindi, Thai, Vietnamese, adding 12 to the previous Preview release. Scores 1.25% error on Seed-TTS, competitive with 4B to 8.5B parameter rivals.
Qwenputer compiles a 32-bit RISC-V CPU directly into Qwen3.5-4B's weights , no external runtime. Gated DeltaNet layers become addressable RAM via one-hot keys with gate and write strength pinned to 1. Benchmarks: 100% on NUPA versus 48.6% base Qwen; 97.5% on DeepMind mul_big versus 55% base.
BKAI's vietnamese-bi-encoder hits 824K+ downloads as a go-to Vietnamese embedding model Built on PhoBERT-base-v2, outputs 768-dim embeddings, ~0.1B parameters, Apache-2.0 licensed Trained on translated MS MARCO, SQuAD v2, and 80% of Zalo 2021 Legal Retrieval challenge Scores 73.28 Acc@1 and 80.73
RustXNG is a Rust rewrite of SearXNG targeted at local AI agent grounding and RAG. Ships as a single binary with no Python or database dependencies, launched via Cargo or Docker. Benchmarks show 0.03s startup vs 0.72s, 13 MiB vs 118 MiB RAM, and 0.27s for 169-engine queries.
EQ-VMamba is the first rotation-equivariant visual Mamba, guaranteeing consistent outputs under 90-degree rotations. Core innovations: EQ-cross-scan with four symmetric paths and group Mamba blocks sharing weights across orientations.
Comfy-Org released a ComfyUI-ready repackage of Speridlabs' Idol image-to-3D model. Idol turns one image into a 3D Gaussian splat with up to 1,048,576 Gaussians. Runs on consumer GPUs in under 7.2 GB of VRAM, generates 1M splats in roughly 3 seconds on high-end hardware.
Haas-ja-v1.0 hits 6.12% CER strict on Japanese ASR, best among open models evaluated. Costs roughly $0.003 per audio hour, about 1/70th of GPT-Transcribe at similar accuracy. Based on NVIDIA Parakeet-TDT with Kimi Delta Attention replacing the LSTM prediction network.
DS Bot is a multi-agent plugin for DeepSeek Harness, now at 4.6k GitHub stars. Ships as a desktop app, DSH plugin, or Docker image; runs fully local under Apache 2.0. Bots have persistent memory, per-bot model selection, and can be configured by chatting with Chief.
Qwen3.5-4B-heretic drops refusals from 99/100 to 13/100 at KL divergence 0.0081 from the base model MMLU, PIQA, and WinoGrande scores stay within one standard error of the original Qwen3.5-4B Produced using Heretic 2.0 with Arbitrary-Rank Ablation on language layers 15 to 27 Ships as 2.71GB Q4_K_M
OhMyGame is an open-source Electron studio where an AI agent designs, builds, and playtests games end to end. Ships for macOS (Apple silicon) and Windows x64 under Apache 2.0, with roughly 1,500 GitHub stars.
Reanimate is an MIT-licensed Claude Code skill that reverse-engineers animations frame by frame. Works from live URLs, mp4, gif, screen recordings, or Lottie files via a headless Playwright browser.
Unsloth released GGUF quants of Qwen-Image-2.1-Turbo, the 8-step distilled version of Qwen's 7B image model Sizes range from 2.47 GB (Q2_K) to 14.23 GB (F16); Q4_K_M at 4.2 GB is the recommended rung Uses Unsloth Dynamic 2.0, which keeps critical layers at higher precision for better accuracy at
Sperid Labs released Idol , an Apache 2.0 image-to-3D Gaussian splat model. 600M parameter 3D-DiT runs on consumer GPUs using 7.2 GB VRAM, outputs standard .ply files. Generates up to 1,048,576 Gaussians per object, 3.4 s on B300, 6.7 s on H100.
DeepSeek open sourced dsh-superagent-prototype , a research prototype of its internal coding super-agent. It is a proactive agent scoped to one GitHub repository, bundled for DeepSeek Harness, MIT licensed.
Mika is an open-source Android AI agent that operates apps on a hidden background screen while you use your phone Over 2,000 GitHub stars within days, Apache-2.0 licensed, written in Kotlin and available on GitHub Bring-your-own-key design supports OpenAI-compatible, Anthropic and Gemini endpoints
Mainplane is a new open-source, MIT-licensed agent harness written in Go. Replaces lossy context compaction with on-demand retrieval from full session transcripts stored as files. Reported 15s and $0.10 to resume a 400k-token session vs 38s and $1.99 for compaction.
NVIDIA released NemotronLabs RE-USE , a 9.6M parameter open multilingual speech enhancement model. Beats 250M and 500M parameter rivals on PESQ, STOI, speaker similarity, and ASR accuracy. Handles noise, reverb, clipping, codec artifacts, packet loss, and bandwidth limits across 7 sample rates.
dsh-Go is a Flutter Android client for the DeepSeek Harness coding agent. Runs DSH locally in a bundled Ubuntu + Node.js environment, no root or Termux required. Can also act as a remote for DSH running on a desktop, over SSH or DSH Remote.
Pydantic V3 development has started, targeting late this year, with no sweeping API rewrite planned. Breaking changes limited to correctness fixes and behaviors you can opt back into, per the V3 tracking issue .
Maya Research released Maya 2 Native , open-weight streaming TTS for 11 Indic languages plus English under Apache-2.0. 2.5B parameters, 20 built-in voices, zero-shot voice cloning from 3 to 10 seconds of reference audio.
Cagliostro Lab released Animagine XL 4.0 , a full SDXL retrain for anime generation. Trained on 8.4M images for ~2650 GPU hours on 7x H100 80GB. Opt variant improves anatomy, color saturation, noise, and output stability.
No stories match these filters.
