Muse Wonder
Meta·
LLMsflagshipcloud
Overview
New model line from Meta Superintelligence Labs, released six weeks after Muse Spark 1.3 together with the larger Muse Wonder Full (separate entry). Available through the Meta Model API and Muse Code at $4 / $20 per million input/output tokens, cached input $0.20. 1M-token context; text, image, video and audio input, text output. Meta's model card reports all results at max reasoning effort and, for the internal benchmarks, notes that competitor runs are best-effort.
Capabilities and innovations
Agentic codingLong-horizon tool useMultimodal input (text, image, video, audio)1M-token contextFirst Muse Wonder releaseLong-conversation memory evaluated with Meta's internal MemoryBench
Benchmarks
| Benchmark | Score | Source |
|---|---|---|
| Humanity’s Last Exam (no tools) | 50.1% | third party HLE 50.1% no-tools, measured by Artificial Analysis at max effort (Intelligence Index v4.3.2 run) (accessed 2026-10-24). |
| Terminal-Bench 4.0 | 49.49% | independent Terminal-Bench 4.0 49.49% (± 2.20), measured by Vals AI at max effort, rank 9 of 54 (accessed 2026-10-24). Artificial Analysis measures 52.0, the figure Meta's model card quotes; Vals AI is used as primary, as for the other recent 4.0 entries. |
- GPQA Diamond: independent evaluation pending — Not on Meta's model card. Artificial Analysis Intelligence Index v4.3 no longer runs GPQA Diamond and Vals AI has archived it (checked 2026-10-24).
- SWE-bench Pro: not reported by the vendor — Meta reports DeepSWE v1.1, SWE-Marathon and ProgramBench instead, as for Muse Spark 1.2 and 1.3.
- MMLU-Pro: not reported by the vendor
- Terminal-Bench 2.x: not reported by the vendor — No Terminal-Bench 2.x run reported. Meta's card quotes Artificial Analysis's 4.0 run, recorded in terminal_bench_4.
- Terminal-Bench 3.0: not reported by the vendor
- DeepSWE v1.1: independent evaluation pending — Meta's model card states its DeepSWE v1.1 figures for Muse Wonder and Muse Wonder Full come from Datacurve third-party runs. Launch coverage quotes the scorecard at 78.8 for this model (79.0 for Muse Wonder Full). The scorecard itself is an image, the card's text carries no number, and Datacurve's public v1.1 board shows no Muse entry, so the figure is not entered until it is read at the source (checked 2026-10-24).
- MMMU-Pro: independent evaluation pending — Artificial Analysis lists no MMMU-Pro value yet (checked 2026-10-24).
API pricing
$4 per million input tokens, $20 per million output tokens (USD, cheapest OpenRouter endpoint, checked 2026-10-24)
Reliability
- omniscience_index
- 25.5 ()independent
Links
More from Meta
Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.