muse-wonder-full

Vendor meta Tier - Price per million tokens, in / out - Released 2026-10-12 Context 1,000,000 Max output 0 Vendor page

Last updated 2026-10-26 How this is made

Meant to excel at

Second-tier-to-top independent results: AA index 56 (tied with Sonnet 5.5, behind only Opus 5.5) and Arena WebDev 1726, rank 6.
Independent evaluator artificialanalysis.ai
Meta's model card reports CyberGym 88.4% and strong refusal rates on bio/chem misuse prompts (about 98 to 100%).
Vendor claim, relayed research.meta.ai

Watch out for

Full variant is limited to Muse Code subscribers with no public API; the API model is Muse Wonder ($4/$20 per 1M), which scores lower (AA 52).
Vendor claim, relayed techcrunch.com
Meta's blog and model card publish few capability scores (blog gives none), so most comparisons rest on third-party indices.
Vendor claim, relayed research.meta.ai
Uses more tokens than average: 140M output tokens on the AA index versus an 89M median.
Independent evaluator artificialanalysis.ai

Benchmarks

Coding percentile 0.89

Benchmark	Score	Percentile	Source	Date
code arena webdev	1726	1.00	Independent evaluator arena.ai	2026-10-19
lmarena code arena webdev	1726	0.78	Independent evaluator arena.ai	2026-10-19

Overall percentile 0.85

Benchmark	Score	Percentile	Source	Date
artificial analysis intelligence index	56	0.85	Independent evaluator artificialanalysis.ai	2026-10-20

Percentile is the share of tracked models this model beats on that benchmark, averaged across the group. Only benchmarks at least four tracked models share are scored.

Practitioner reports

+

A short default-settings test of Muse Wonder (not the Full variant) passed 29 checks: fixed a JS dedupe function correctly, resolved conflicting sources into valid JSON listing gaps, and wrote a six-step recovery plan.

2026-10-14
Practitioner report glbgpt.com