Meta's Muse Spark 1.3: 38% fewer errors than Opus 5 at $0.85 per correct task
ryanshrout · x · 2026-09-04
Signal65's Ryan Shrout published new PINNACLE agentic benchmark results: Meta's Muse Spark 1.3 ranks second, with 38% fewer weighted errors than Claude Opus 5, and costs $0.85 per correct task versus $1.36 for Opus and $1.22 for GPT-5.6 Sol — knocking both hosted frontier models off the cost frontier.
Details:
- The model is slow and token-hungry, yet still the best buy at the top of the hosted tier
- PINNACLE builds real multi-step enterprise tasks from US Department of Labor work activities, with deterministic code-verified scoring, no model judges, and a freshly generated sandbox and answer key each run to prevent memorization
- Full results are available on the Signal65 site
This may be Meta's next "Llama moment."
More from Models
- Open models below 4-bit struggle at self-iterating 3D asset generation, HF dev finds — mervenoyann · 2026-10-21
- Researcher Finds Two Windows 0-Days, First in 76 Minutes with a Frontier Model — moyix · 2026-10-21
- Dev finds Gemini Argon's debug-log analysis beats Fable 5.1 — ShikharMurty · 2026-10-21
- ChatGPT Gets Stuck Screaming "AAHH" in an Infinite Loop After a Simple Prompt — AutomaticScene · 2026-10-21
- Open-weights Ornith-1.5-122B debuts with 84.5% SWE-bench Verified, runs on a single DGX card — _akhaliq · 2026-10-21
- qev-0.8B: a tiny local decision model that picks moves in 3.5ms on 1.3GB — fajfas3 · 2026-10-21