Last 90 days · 2711 total
Ollama's v0.30.9 release introduces support for the Cohere2Moe architecture and resolves issues with LFM2 parser/renderer and coding agent output.
OriginalA Reddit discussion highlights the urgent need for larger 80-160B language models optimized for unified memory devices like Apple or Ryzen AI with >9…
OriginalClaude Code v2.1.178 introduces new Tool(param:value) syntax for fine-grained permission rules and improved loading of skills from nested directories.
OriginalvLLM has released version 0.23.0, featuring 408 commits from 200 contributors, though Minimax M3 support is not yet included.
Originalllama.cpp's b9690 update refactors the Metal `rope_back` operator to reuse existing kernels, enhancing efficiency for macOS/iOS.
OriginalWilson is an AI coworker integrated into Slack that can build reports, create work tools, and perform various other tasks.
OriginalGLM-5.2, a massive 753B parameter, MIT-licensed coding agent, is celebrated as a significant advancement for local AI, despite its high hardware requ…
OriginalGLM-5.2 has been recognized as the top-performing open weights model on the Artificial Analysis intelligence index.
OriginalInflect-Nano-v1, an ultra-compact 4.63 million parameter Text-to-Speech (TTS) model, has been released, demonstrating surprising performance for its…
OriginalA reminder from Tibo emphasizes that the Codex App, CLI, and SDK are designed to work with any open source model, not exclusively OpenAI's.
Original