Meta's Muse Spark 1.3 tops coding benchmarks vs GPT 5.6 and Opus 5, open weights coming
armand_ruiz · x · 2026-09-03
- Meta shipped Muse Spark 1.3, claiming its biggest jump yet on coding and agentic work at near-commodity pricing, live in Muse Code and the API.
- Third-party benchmarks show strong coding results: DeepSWE v1.1 at 75.4 beats GPT 5.6 Sol (73.0) and Opus 5 (74.0); codebase understanding (SWEAtlas) 59.4 clears both by 6+ points; Terminal-Bench 2.1 at 88.8 ties GPT and beats Opus's 86.
- Meta also teased an upcoming model (watermelon codename) and open-weight Muse Spark releases.
Related event: Meta Launches Muse Spark 1.3: Near-Frontier Performance at ~1/8 the Price(53 posts)→
More from Models
- Opus 5.5 tops OpenRouter spending in 20 of 29 task categories — mmurph · 2026-10-21
- Meta team builds open-source 30B LLM from scratch in months, hails 'second era' — TimDarcet · 2026-10-21
- DFlash drafter for Muse Glimmer-30B: 5 layers, blockwise 16-token spec decoding — TimDarcet · 2026-10-21
- EvoPress quantization shrinks 30B model from 56GB to 19GB at near-bf16 accuracy — TimDarcet · 2026-10-21
- Radar chart: Muse Glimmer-30B benchmarked against Gemma4-31B and Qwen3.6-27B — TimDarcet · 2026-10-21
- Privacy evals: Muse Glimmer-30B leaks less than Qwen3.6-27B on CIMemories — TimDarcet · 2026-10-21