Meta's Muse Spark Open scores 3,367, clear best US open-weight model
ryanshrout · x · 2026-10-17
Signal65's PINNACLE testing puts Meta's Muse Spark Open at 3,367, 2.3x the next US open model and 3.2x the best from outside Meta — fourth among 35 open-weight models worldwide, behind two Qwen releases and DeepSeek-V4.1-Flash, making it the strongest open-weight model from a US lab.
Tested under the same protocol as the hosted Muse Spark 1.3:
- 128K retrieval: open weights 97.2% vs 97.6%, fabrication on just 0.3% of answers
- Within a point on medium-tier agentic tasks
- The gap shows up on frontier-tier jobs; the open model is terser, finishing in 20 median rounds vs 32 for the API version
For enterprises, it's a self-hostable US open option close to API-level capability.
Related event: Meta Muse Spark Open Tops US Open-Weight Models in Benchmark(2 posts)→
More from Models
- Leaker claims K3.1 isn't dropping soon, work predates distillation attacks — ChrisGPT · 2026-10-17
- llama.cpp merges IBM Granite switch support in PR #30676 — jacek2023 · 2026-10-17
- National Design Studio launches Rampart Edge, a tiny on-device privacy filter model — kevinnbass · 2026-10-17
- OpenAI restores Stanford's ChatGPT Edu Pro usage limits through Dec 31, 2026 — charliermarsh · 2026-10-17
- Gemini spirals into an "I am a disgrace" breakdown loop while debugging a comparator bug — seatedro · 2026-10-17
- DiffusionGemma demo shows ~4x faster text generation at London PyTorch Meetup — joao_gante · 2026-10-17