[{"data":1,"prerenderedAt":28},["ShallowReactive",2],{"nr-en-meta-muse-spark-open-816b-open-weight-model":3},{"slug":4,"title":5,"dek":6,"date":7,"time":8,"publishedAt":9,"updated":10,"updatedAt":10,"dateFmt":11,"updatedFmt":10,"kind":12,"tier":13,"author":14,"authorName":15,"topics":16,"tracker":20,"trackerLabel":21,"headlineStat":22,"image":23,"ogImage":24,"imageAlt":5,"csv":10,"minutes":25,"words":26,"html":27},"meta-muse-spark-open-816b-open-weight-model","Meta releases Muse Spark Open: per Meta, the most capable US open-weight model with 816 billion parameters","Meta returns with an open model that, by its own account, outperforms other open models but trails paid models. The weights are available on Hugging Face.","2026-10-16","07:03","2026-10-16T07:03:00+02:00","","October 16, 2026","news","standard","ideal-syka","Ideal Syka",[17,18,19],"Open-weight models","Meta","Local AI","\u002Flokale-ki","Local AI models","816 billion parameters (42 billion active)","\u002Fnewsroom\u002Fimg\u002Fmeta-muse-spark-open-816b-open-weight-model.webp","\u002Fog-nr\u002Fmeta-muse-spark-open-816b-open-weight-model.en.png",1,225,"\u003Cp>According to Meta, Muse Spark Open is the most capable US open-weight model to date, designed for programming and agentic tasks. The source The Decoder (DE) reports 816 billion parameters, 42 billion of them active, up to one million tokens of context, and support for text, image, video and audio.\u003C\u002Fp>\n\u003Ch2>Benchmarks with caveats\u003C\u002Fh2>\n\u003Cp>In benchmarks such as JobBench, MCP-Atlas, HLE-Diamond and ZeroBench, the model is according to Meta in some cases clearly ahead of Qwen 3.8, GLM 5.3 and Kimi K3, but clearly behind the paid models Muse Wonder Full. According to Meta, it is in some coding benchmarks behind the competition.\u003C\u002Fp>\n\u003Ch2>License and operation\u003C\u002Fh2>\n\u003Cp>The license allows free use but requires the notice "Built with Muse Spark Open". According to The Decoder, providers with over 20 million dollars in annual revenue that offer the model as an API service need a separate agreement with Meta. For local operation, Meta cites two Mac Studios with 512 GB of memory each as sufficient.\u003C\u002Fp>\n\u003Ch2>Classification\u003C\u002Fh2>\n\u003Cp>For German companies, an open model of this size may be interesting if data should stay local. Whether the performance holds up in a specific use case cannot be judged from vendor figures alone.\u003C\u002Fp>\n\u003Ch2>Sources\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fthe-decoder.de\u002Fmuse-spark-open-meta-meldet-sich-mit-open-weight-modell-optimiert-fuer-apples-m5-zurueck\u002F\">The Decoder (DE)\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fresearch.meta.ai\u002Fblog\u002Fmuse-spark-open\">Meta AI\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cem>Editorially owned by \u003Ca href=\"\u002Fen\u002Fautor\u002Fideal-syka\">Ideal Syka\u003C\u002Fa>. Sources and method: \u003Ca href=\"\u002Fen\u002Fredaktion\">Newsroom & method\u003C\u002Fa>. Tips and corrections: \u003Ca href=\"mailto:ai@i6eal.de\">ai@i6eal.de\u003C\u002Fa>.\u003C\u002Fem>\u003C\u002Fp>\n",1792135279390]