Meta ships MuseSpark 1.3 hours after Google, overtakes Gemini 3.8 Flash on AI index
智东西 · wechat · 2026-09-03
Four hours after Google's Gemini 3.8 Flash launch, Meta released MuseSpark 1.3, scoring 62 on the Artificial Analysis index to edge past Gemini's 59, behind only Claude Fable 5.1 (66) and Claude Opus 5 (63). Meta's chief AI officer Alexandr Wang taunted "gemini who?" on X. It's Meta's fourth MuseSpark model in five months, with open-weight releases promised next.
The model targets long-context and coding—MuseCode + MuseSpark 1.3 rivals Claude Code + Opus 5—and cuts tool calls by 20% and token usage by 25% versus 1.2, with better constraint retention in multi-step tasks and improved robustness against prompt injection. Pricing is unchanged: $1.25/M input tokens (cache miss), $4.25/M output—cheaper than DeepSeek-V4-Pro's peak rates.
Developer reception is mixed: fast and cheap but not stunning. One built Minecraft for 10 cents; another ran an Ultra-mode session with 20 self-optimization loops and 60+ agent tasks in 2 hours for under $1. But on the MineBench 3D reasoning eval it cost 5.6x more than Gemini 3.8 Flash, ran slower, and scored lower, and agentic web-search tasks remain a weak spot.
Related event: Meta launches Muse Spark 1.3, matching Fable 5 at a fraction of the price(53 posts)→
More from Models
- GPT-6 Astra and Astra Aeon spotted in Codex; Aeon tipped as long-horizon agent model — VraserX · 2026-09-03
- Ant Group Open-Sources Finance-Enhanced Model Ling-3.0-flash-Fin with 124B Parameters — niacolhealth · 2026-09-03
- Should 'hard scientific problems solved' be the new LLM benchmark? — Dr_Singularity · 2026-09-03
- Ben Thompson on Gemini: 'OK, it's now or never' — kieranklaassen · 2026-09-03
- User Reports ChatGPT Randomly Outputs 'GOD IS COMING' and Won't Stop — jaypro1005 · 2026-09-03
- Ling-3.0-flash-Fin Weights Released: 124B Params, 5.1B Active, 256K Context — Bestlife73 · 2026-09-03