Meta ships Muse Spark 1.3: 98.5 on MRCR 256K, 25% fewer tokens, near Opus 5 on agentic benchmarks
rohanpaul_ai · x · 2026-09-03
Meta released Muse Spark 1.3, calling it its biggest jump on coding and agentic work, priced "almost too cheap to meter" and live in Muse Code and the API; open weights and a next release are teased. Key numbers: 20% fewer tool calls and 25% fewer tokens than 1.2, 98.5 on MRCR 256K–512K (beating GPT-5.6 Sol), and near-Opus 5 results on JobBench (64.9 vs 65.7), OSWorld (66.9 vs 68.3), and AutomationBench (49.4 vs 50.3). Vendor-reported figures suggest coding work will need fewer model actions and less generated text.
Related event: Meta launches Muse Spark 1.3, matching Fable 5 at a fraction of the price(53 posts)→
More from coding & agent
- A roundup of open-source AI agent platforms: specialized agents, scheduling, MCP tools — resiros · 2026-09-03
- Debugging multi-step LLM workflows: how developers trace failures upstream — Dry_Raise_3418 · 2026-09-03
- Solo.io open-sources agentdesktop, a visibility layer to manage AI agents across desktop fleets — pjausovec · 2026-09-03
- Unify CTO's rule: your LLM judge must run on a different model family, or agents collapse into groupthink — LangChain · 2026-09-03
- Meta's CORAL puts AI agents in charge of a production recommender serving billions, with real A/B results — omarsar0 · 2026-09-03
- Is Obsidian just a graph of Markdown? Designing an AI-native second brain — NyeinChanSoe · 2026-09-03