ByteDance's Seed 2.1 Pro Preview debuts at #8 on Arena's frontend code leaderboard, matching Claude Opus 4.6 on React and UI tasks before public release.
CodeBurn is a free, local-only TUI that breaks down every token and dollar you spend across Claude Code, Codex, Cursor, and more — with waste detection built in.
Cognee is an open-source graph-vector memory engine that gives AI agents persistent, relationship-aware recall across sessions in under 10 lines of code.
GitHub and Microsoft cut secret scanning false positives by 75.76% using an LLM that reads how code actually uses a detected value, not just what it looks like.
Datalab open-sources a 9B vision model that extracts structured JSON from documents at near-frontier accuracy, running 3x faster than Gemini Flash 3.5.
Allen AI's OLMo-core is the fully open PyTorch training framework behind OLMo 2 and OLMo 3, now supporting MoE, multimodal, and 4-way parallelism out of the box.
PrimeIntellect's open-source Verifiers library turns RL environments into shareable Python packages, powering everything from quick evals to 100B+ model training runs
Megatron Core's full development is now open on GitHub, with a new MoE technical report, Dynamic Context Parallelism, and benchmarks scaling to 462B parameters on 6,144 H100s.
GitHub's official MCP server gives any AI agent full programmatic access to repos, issues, PRs, Actions, and security tooling -- with nearly 30k stars and growing fast
SentrySearch lets you type a plain-English query and instantly extract the matching clip from hours of raw dashcam footage — no transcription, no tagging required.
Tencent's AngelSlim open-sources a full LLM compression stack covering quantization down to 1.25-bit, Eagle3 speculative decoding, and sparse attention — all in one pip install.
GitHub's new standalone desktop app lets you run and manage multiple AI coding agents in parallel, from issue to merged PR, without leaving one surface.
Microsoft's FastContext offloads repo search to a tiny 4B model, cutting main-agent token use by up to 60% while improving benchmark scores across GPT-5.4, GLM-5.1, and Kimi-K2.6.
Datalab open-sources lift, a 9B vision model that extracts structured JSON from PDFs at near-frontier accuracy with 3x faster latency than Gemini Flash 3.5