Strengths
- A clear step up from Muse Spark 1.3: 52 instead of 48 on the Artificial Analysis index, and 52% on the Terminal-Bench 4.0 terminal work test.
- It rarely makes things up: its hallucination rate in Artificial Analysis testing is 27%, and 23% for Wonder Full, the second lowest among models scoring 45 or more.
- A 1M-token window with no surcharge for long inputs. It understands text, images, video, audio and PDFs.
- Meta says it is built for long agent work: it plans, hands tasks to helper agents, calls several tools at once and can operate a computer.
Weaknesses
- It is expensive: one Artificial Analysis task costs about $4.52, almost three times as much as Muse Spark 1.3.
- Its general knowledge is weaker: 25 on AA-Omniscience, against 46 for Claude Opus 5.5 and 43 for GPT-6 Astra.
- Little transparency: Meta does not disclose its knowledge cutoff, parameter count or training details.
- It is still in preview, and the strongest version, Wonder Full, is not on the API at all.
Best for
- code changes across many files
- agents that run for hours
- coordinating helper agents
- operating a computer
Not ideal for
- large volumes of cheap tasks, Muse Spark is better for that
- fact-heavy questions without web search
How to use it in Europe
Muse Wonder is not part of the free Meta AI assistant. You use it through the Meta Model API or in a Muse Code subscription from $5 a month, a command-line tool for developers. Meta's rules do not exclude Europe, and South Korea is the only excluded country. Meta does not publish an official country list, though, and does not mention any option to process data in the EU.
What is coming
Confirmed by vendor Meta has announced that Muse Wonder, one of its two most capable models, will soon power features in the Muse app as well. Subscribers get it first, and Meta has not given a date. When: no date yet, Meta says “soon”, subscribers first →
Price compared with similar models
Benchmark results
| Benchmark | Score | Source |
|---|---|---|
| Artificial Analysis Intelligence Index v4.3 (max) | 52 (Muse Wonder Full: 56) | artificialanalysis.ai ↗, October 2026 |
| Terminal-Bench 4.0 (AA) | 52% (Muse Wonder Full: 57%) | artificialanalysis.ai ↗, October 2026 |
| Artificial Analysis Coding Agents Index (Muse Code) | 63 (Muse Wonder Full: 68) | artificialanalysis.ai ↗, October 2026 |
| AA hallucination rate (lower is better) | 27% (Muse Wonder Full: 23%) | artificialanalysis.ai ↗, October 2026 |
Benchmarks cover only a narrow slice of ability and vendors often report them at their best settings. Treat them as a rough guide.
More models from this vendor
- Muse Spark 1.3: Meta's balanced model with a 1-million-token window and a fair price. The Muse Spark line also powers the Meta AI assistant.
- Muse Spark Open: Meta's large open model that a company can run on its own servers, so data never leaves its infrastructure.
- Muse Glimmer 30B: Meta's small open model under Apache 2.0 that runs on a single graphics card, even offline.
- Llama 4 Maverick: The larger model in the newest Llama generation, from April 2025. Cheap through providers, but its license excludes people and companies in the EU.
- Llama 4 Scout: The smaller Llama 4 from April 2025, very cheap through providers. Its license grants no rights to people or companies in the EU.
Similar models from other vendors
- Claude Opus 5.5 (Anthropic): Anthropic's main model for demanding work, coding and long documents, and the one Anthropic recommends starting with.
- DeepSeek V4 Pro (DeepSeek): DeepSeek's largest open-weight model for demanding text tasks at a low price. It does not understand images.
- Gemini 4 Carbon (Google): Google's most capable model, announced on 28 October 2026. So far only selected cybersecurity experts have access.
- Gemini 4 Argon (Google): Google's most capable publicly available model, out in October 2026. Built for demanding analytical work where accuracy matters and made-up answers are costly.
- Mistral Large 4 (Mistral AI): Mistral's largest model, still in preview. Built for demanding tasks, long documents and security analysis with data kept in the EU.
- Mistral Large 3 (Mistral AI): Mistral's previous large model with open weights under Apache 2.0. Cheap in the API and suited to your own powerful server.