Meta ships Muse Wonder at Opus 5.5's $4 / $20, and says Wonder Full beats Opus 5.5 on 8 of 14 benchmarks

Meta released Muse Wonder today, its most capable model and a new line above Muse Spark. Muse Wonder is on Meta Model API now as muse-wonder-1.0, at $4 per million input tokens and $20 per million output, the same price as Claude Opus 5.5. A stronger variant, Muse Wonder Full, is only in Muse Code, on the Power Usage and Maximum Usage plans. Meta announced both on its research blog on October 12, with API docs and a model card.

THE PRICE AND SPECS

- $4 input / $20 output per 1M tokens, $0.20 for cached input. Meta charges no long-context premium. - 1,048,576-token context window and 131,072 tokens of max output (the output limit is from models.dev). - Text, image, video, audio and PDF in; text out. Reasoning effort runs from "minimal" to "max". - The API model is labelled preview, and it is Standard mode only: the discounted Contributor mode, which lets Meta train on your data, is not offered for Wonder on pay-as-you-go. - For scale, Muse Spark 1.3 costs $1.25 / $4.25.

META'S OWN BENCHMARKS

These are Meta's figures, run at max reasoning effort against Claude Opus 5.5, Claude Fable 5.1 and GPT-6 Astra:

- Muse Wonder Full leads Opus 5.5 on 8 of the 14 rows in Meta's table. Two of those 8 are Meta's own internal tests: its Personal Superintelligence Index and MemoryBench. - The biggest coding leads: DeepSWE 1.1, 79.0 vs 74.2, and SWE-Marathon, 57.5 vs 46.3 (Wonder Full at xhigh effort). - Opus 5.5 stays ahead on Terminal-Bench 4.0 (59.6 vs 57.1), ProgramBench (70.5 vs 59.0), HealthBench Pro (65.6 vs 62.3), JobBench (by 0.1) and the partial OSWorld V2.1 score. - Plain Muse Wonder, the model on the API, scores below Wonder Full on most rows.

THE INDEPENDENT NUMBER

Artificial Analysis scores Muse Wonder 52 and Muse Wonder Full 56 on its Intelligence Index (read on its site today). On the same index, as Meta's chart shows it, Claude Opus 5.5 is at 58, Claude Fable 5.1, GPT-6 Astra and Gemini 4 Argon at 53, and GPT-6.1 Sol at 52. So among the models in Meta's chart, Opus 5.5 is still first and Wonder Full second, and the model you can call on the API sits level with GPT-6.1 Sol.

SAFETY

Meta's model card says that without mitigations it does not rule out Wonder Full reaching its high-risk threshold for chemical and biological misuse, so the model ships with safety training and classifiers. Meta rates cyber and loss of control "moderate or lower". A full safety report is due "soon".

WHAT WE DO NOT KNOW

- Whether Muse Wonder Full will come to the API, and at what price. - How it does on the other public boards. Only Artificial Analysis has a score so far. - Whether this is the model reported as Watermelon. Meta's post does not use the codename.