Meta releases Muse Spark Open, an 816B-parameter open-weight multimodal model
Meta released Muse Spark Open, an 816B-parameter multimodal model that can run locally on Apple Mac Studios. Two M5 Ultra Mac Studios with 1TB combined memory run the model via Apple’s updated MLX toolkit, and Meta says it shipped iOS app fixes and edited video in DaVinci Resolve without cloud.
Key points
- Meta releases 816B-parameter Muse Spark Open, an open-weight multimodal model
- Two M5 Ultra Mac Studios with 1TB memory run Spark Open via Apple MLX toolkit
- Model ranks 39 on Intelligence Index, ahead of West open models but behind China’s GLM 5.3, Kimi K3, Qwen3.8
On the Artificial Analysis Intelligence Index, the model sits at 39, ahead of West open-weight models but behind China’s GLM 5.3 (45), Kimi K3 (44) and Qwen3.8 (40). Spark Open matches or tops open rivals on multimodal and knowledge tasks, though Chinese models still code better. It is the third major Western open launch this month after Reflection AI’s Beam and Mistral’s 1T-parameter Le Chonk, whose weights arrive Oct. 27.
Meta’s move signals a return to open model strategy after its earlier shift to closed Llama models. The release gives U.S. developers a large open-weight model that can run locally, potentially reducing cloud dependence, and challenges China’s dominance in the open-model race.
The story so far
4 episodes →- Meta releases Muse Spark Open, an 816B-parameter open-weight multimodal modelthis story
Meta turns Mac into an AI supercomputer
The Rundown AI · 16 October 2026
Loading the full article…
This text was published by The Rundown AI and written by Zach Mink. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- DeepSeek’s Flash ties Anthropic’s Opus 5.5 in browser-agent benchmark · 2 src
- MiniMax releases M3.1 Flash model at $0.15 per 1M input tokens · 4 src
- Mercury 3 [Inception Labs] - Matching Luna/Haiku at 1000 tokens/sec · 2 src
- Kling AI releases version 4.0 with stereo audio and a lighter Flash variant · 1 src
- DeepSeek releases V4.1-Flash-1017 update with lower cost and higher DeepSWE score · 26 src
Comments
via GitHub Discussions