Quick Answer
Muse Spark 1.1 vs Opus 5 vs GPT-5.6 Sol: Agents
The Short Answer
Muse Spark 1.1 wins on price and some agentic tool-use benchmarks; Claude Opus 5 and GPT-5.6 Sol win on overall intelligence and proven reliability. Meta shipped Muse Spark 1.1 on July 9, 2026 as its first paid model — $1.25/$4.25 per MTok, roughly a quarter the price of the frontier leaders.
Pricing (the headline)
| Model | Input | Output | 30K-in/5K-out task |
|---|---|---|---|
| Muse Spark 1.1 | $1.25 | $4.25 | ~$0.059 |
| Claude Opus 5 | $5 | $25 | ~$0.28 |
| GPT-5.6 Sol | $5 | $30 | ~$0.30 |
Agentic Benchmarks (Meta’s numbers)
Meta’s evals put Muse Spark 1.1 ahead of Opus 4.8 and GPT-5.5 on tool-use:
- MCP Atlas (tool use at scale): 88.1 vs Opus 4.8 82.2, GPT-5.5 75.3
- JobBench (professional tool-use): 54.7 vs Opus 4.8 48.4, GPT-5.5 38.3
- Finance Agent v2: 57.2 vs Opus 4.8 53.9, GPT-5.5 51.8
Caveat: these are vendor-run against Opus 4.8/GPT-5.5, not Opus 5 or GPT-5.6 Sol.
Overall Intelligence & Coding
- Artificial Analysis Intelligence Index: Muse Spark 1.1 = 51 (up from 1.0’s 43), near GPT-5.6 Luna max / GLM-5.2 max — still behind Claude Fable 5, GPT-5.6 Sol, and Claude Opus 5.
- SWE-bench Pro: Muse Spark 1.1 61.5 · Claude Opus 5 79.2% · GPT-5.6 Sol 64.6%.
- Terminal-Bench 2.1: Muse Spark 1.1 80.0 (contested on HN over resource-cap overrides) · GPT-5.6 Sol Ultra ~91.9%.
What to Do
- Reliability-first agents: default to Claude Opus 5 (best multi-file coding + long-horizon) or GPT-5.6 Sol (terminal-heavy execution).
- Cost-sensitive, high-volume agents: route Muse Spark 1.1 — its tool-use scores are strong and it’s ~4x cheaper.
- Keep the model a config value so you can A/B per task type.
Sources
- Meta AI — Introducing Muse Spark & Meta Model API: ai.meta.com/blog/introducing-muse-spark-meta-model-api
- Artificial Analysis — Muse Spark 1.1: artificialanalysis.ai/models/muse-spark-1-1
- Anthropic — Claude Opus 5 (July 24, 2026): anthropic.com/news/claude-opus-5