Home · 2

Mistral Large 4 features a 1T Mixture-of-Experts architecture, 1M context, and top scores on coding and security benchmarks.

Background status updates in Claude Code consume nearly a fifth of model usage. A Bun preload hook suppresses these requests to save rate limits.

Liquid AI introduces LFM3, a multimodal family built on a shared 1.7B-parameter backbone. Includes text, vision, audio, and omni models optimized for edge deplo

Claude Code v2.1.311 fixes 200K token compaction errors, improves subagent MCP connectivity, and adds hold parameters for self-hosted runners.

Claude Code v2.1.312 brings custom OIDC user ID claims, concurrency limits on execution hooks, and safeguards against truncating large project files.

Microsoft and Hugging Face release ThinkingBox-Training, using executable database checks as GRPO rewards. Qwen3.8-27B jumps from 7.5% to 21.1% observed 20/20 r

IBM updates its Granite Speech collection with the 2B parameter 5.0 model, a 19,870 RTFx Turbo CTC encoder, and improved 4.1 2B Plus weights.

Ollama v0.40.11 adds local macOS speech generation via an OpenAI-compatible endpoint and integrates d3 decision models in 2B to 27B sizes.

n8n and Cognee demonstrate how hybrid knowledge graph architectures solve state persistence for production AI workflows.

Claude Code v2.1.309 patches sandbox leaks, fixes broken transcript JSON from emoji truncation, and adds namespacing for directory skills.