AI Briefing
KOSign in

Meta's 816B Muse Spark Open scores 39 on Artificial Analysis, trailing 180B Qwen3.8-Flash-Next

·2026.10.23 05:14

Key point

The 816B parameter model scores 39, slightly below the 180B Qwen3.8-Flash-Next which achieves a score of 40.

Details

  • Meta's 816B Muse Spark Open (42B active) scores 39 on Artificial Analysis, while the 180B Qwen3.8-Flash-Next (6B active) scores 40.
  • Unsloth UD-Q4_K_XL GGUFs require approximately 506 GB for Muse Spark Open and 111 GB for Qwen3.8-Flash-Next.
  • Meta recommends a local setup of two M5 Ultra Mac Studios with 512 GB each for Muse Spark Open.
  • Qwen3.8 27B scores 34, outperforming Beam (501B) at 30 and Nemotron 3.5 Bolt (74B) at 21, with 4-bit quants requiring 16 to 19 GB of memory.
  • MiMo-V2.6-Pro (1T) leads the chart with a score of 46, followed by DeepSeek V4.1 Flash 1017 (552B) and GLM-5.3 (753B) at 45.
  • The Artificial Analysis index is a single aggregate number that does not reflect speed, long context, vision, or specific task performance.
benchmarkqwenlocal-llmmetaartificial-analysisopen-weight-models
Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.

Related

RedditNace.ai releases Kapitsa, an open-source Apple Silicon inference engine for Qwen and NemotronRedditAlibaba Qwen team releases BlenderWorld benchmark for 3D scene creation agentsGeekNewsMeta revises commercial terms to remove 'train' but retains broad scraping rights for AI business productsPyTorchKRStrata Engine Released to Run Qwen3.8-Flash-Next 125B on 12GB VRAMHuggingFace BlogMicrosoft Releases ThinkingBox Training Stack for RL Post-Training of Agents Using Database State RewardsNVIDIA BlogEuropean Financial Institutions Deploy Open Models on NVIDIA Infrastructure for Proprietary Intelligence