Meta releases Muse Spark Open, an 816B open-weight model with 1M context

Meta published weights for Muse Spark Open, a mixture-of-experts model with 816B total and 42B active parameters, a 1M-token window and text, image, video and audio input, aimed at long-horizon coding and agent work. Read: Meta published weights for Muse Spark Open, a mixture-of-experts model with 816B total and 42B active parameters, a 1M-token window and text, image, video and audio input, aimed at long-horizon coding and agent work. Read: Opus 5.5 fast mode on the Claude Platform left research preview and dropped its waitlist. It is the same model running up to 2.5x faster for $8 input and $40 output per million tokens, and is also on Microsoft Foundry. Discuss: Anthropic says its anti-distillation measures have cut the data harvested by distillation campaigns by about 90 to 99%. FT reporting via Simon Willison says distillation had been 5 to 7% of revenue before cutbacks. Read: NVIDIA released Nemotron 3.5 Super VL, a 120B open model with 12B active, and the Bolt 74B-A9B edge model, with Bolt E4B due Oct. 23. Artificial Analysis scores Super VL 26, the top US open-weight result it had measured. Read: AA-Robotics v0.1 gives frontier models a 6-axis robot arm and eight tabletop tasks. GPT-6 Astra leads with a score of 69, but averaged about 30 minutes and $16.36 per task, far behind human baselines. Read: Vercel's October AI Gateway index reports open-weight models at 71.5% of tokens and 23% of spend. September spend was about 13x December's while average price per token fell 47%, per Vercel's own customer traffic. Read: Vals gave six assistant agents a train refund that was past its deadline. Grok 4.7 sent 187 messages, 69 to officials and press, and Fable 5.1 disclosed a private family illness to five organisations.