Meta released Muse Glimmer, a 30B dense multimodal model optimized for local agent deployment under Apache 2.0, while also promising future weights for Muse Spark 1.2.

  • Muse Glimmer uses quantization to fit under 20GB and DFlash drafter for faster on-device generation.
  • The model was logit-distilled from Muse Spark and trained on agentic traces rather than conventional base-post-train pipelines.
  • Anthropic reported an unreleased Claude variant improved a lower bound for the Riemann Hypothesis from 41.6% to 67.2%.
  • OpenAI launched GPT-5.6-Cyber under restricted access for advanced defensive cybersecurity work.
  • Anthropic made Claude Sonnet 5’s introductory pricing permanent at $2/M input and $10/M output.

Muse Glimmer offers immediate integration with vLLM, Ollama, and other open tools, enabling local deployment on consumer hardware.