Muse Wonder for Chemical Companies: The Top Version Only Runs in Muse Code
Meta's Muse Wonder ranks 4th of 27 on real professional work, at the same token price as Claude Opus 5.5. Muse Wonder Full ranks 2nd, level with Opus, but Meta offers it only inside Muse Code, its coding tool, so a chemical distributor or producer can't build it into its own systems.
Published
First published.
Muse Wonder at a glance
- Made by
- Meta7
- Released
- 12 Oct 20263
- Price
- $4 per million tokens sent and $20 per million written, the same as Claude Opus 5.51
- Where to use it
- Muse Wonder on the Meta Model API and in Muse Code; Muse Wonder Full in Muse Code only6
- Reads
- Up to about 1 million tokens at once3
- Weights
- Closed: you can't run it on your own servers3
What Muse Wonder is.
Meta released Muse Wonder and Muse Wonder Full on 12 October 2026. They are general AI models, the kind that write, summarize, and work through a long task, and Meta calls them its most capable so far. Meta built them for agents, work that runs over many steps, with what it calls Active Memory, carrying details across conversations, and Active Focus, keeping at a task when it gets hard.
Muse Wonder is on the Meta Model API, so your own software can call it, and in Muse Code, Meta's coding tool. Muse Wonder Full is available only to Muse Code subscribers. Neither is open: you can't download them and run them on your own servers.
Two of the guide's tests scored both versions on the day they came out. Artificial Analysis ran each at its highest setting.
Muse Wonder and Muse Wonder Full, test by test.
Meta released two versions on the same day. Each row is a test that scored both, and each track runs from first place on the left to last on the right. The filled dot is Muse Wonder, the version your software can call; the ring is Muse Wonder Full.
Muse Wonder
On the APIOn the Meta Model API and in Muse Code
$4 per million tokens sent, $20 per million written
Muse Wonder Full
Not on the APIIn Muse Code, Meta's coding tool, for subscribers only
No API price
Finished professional work
Artificial Analysis, 27 models ranked on real professional tasks
Artificial Analysis: GDPval-AANo date stated
Regulatory research
vals.ai Legal Research, 78 models, questions answered fully right
vals.ai Legal Research, regulatory questions12 Oct 2026
Muse Wonder compared with Claude Opus 5.5 and Sonnet 5.5.
The two models it meets most often on price. All three are inside the regulatory test's margin of error, so on that job the difference is the bill.
| Muse Wonder | Claude Opus 5.5 | Claude Sonnet 5.5 | |
|---|---|---|---|
| Price per million tokens sent and written | $4 and $20 | $4 and $20 | $2 and $10 |
| Cost of one index task | $4.52 | $5.98 | $5.46 |
| Real professional work | 4th of 27 | 1st of 27 | 3rd of 27 |
| Regulatory research, fully right | 63.64%, $3.92 a question | 65.15%, $22.21 a question | 69.70%, $12.64 a question |
Artificial Analysis: models and the Intelligence IndexNo date stated
Artificial Analysis: GDPval-AANo date stated
vals.ai Legal Research, regulatory questions12 Oct 2026
Where Muse Wonder falls short.
What the tests and Meta's own pages show, before you put it to work.
The Top Version Only Runs in Muse Code
Any company can subscribe to Muse Code, Meta's coding tool, and use Muse Wonder Full there. It isn't on the API, though, so you can't build it into your own systems, such as one that reads your order emails. Those can call only Muse Wonder, 4th of 27.
Meta AI Research: Muse Wonder12 Oct 2026
No Document Test Yet
Neither OmniDocBench nor ParseBench has scored it, so how it reads your TDSs, SDSs, and CoAs is untested. Try it on your own files first.
OmniDocBench leaderboard11 Sep 2026
Meta's Own Results Are Meta's
Meta says the two versions lead on personal and professional agent work, memory, and coding. Those are Meta's own tests. On the independent ones here, the version you can call ranks 4th.
Meta AI Research: Muse Wonder12 Oct 2026
A US Regulatory Test
The regulatory test asks 66 questions on US administrative law, not REACH, CLP, or GHS. Check its answers on your own rules.
vals.ai Legal Research, regulatory questions12 Oct 2026
It Runs on Meta's Servers
Its weights are closed, so you can't run it on your own computers. Every document you send it goes to Meta's service.
Artificial Analysis: Muse WonderNo date stated
Questions about Muse Wonder.
Short answers, from the same sources as the rest of this page.
What is Muse Wonder?
Meta's most capable general AI model so far, released on 12 October 2026 with a larger version, Muse Wonder Full. Meta built both for long, multi-step work.
What is the difference between Muse Wonder and Muse Wonder Full?
Muse Wonder Full ranks 2nd of 27 on real professional work, level with Claude Opus 5.5, and Muse Wonder ranks 4th. Muse Wonder is on the Meta Model API and in Muse Code; Muse Wonder Full is only in Muse Code, for subscribers.
How much does Muse Wonder cost?
$4 for every million tokens sent and $20 for every million written, the same as Claude Opus 5.5. One task on the Artificial Analysis Intelligence Index cost $4.52, against $5.98 on Opus 5.5.
Muse Wonder or Claude Opus 5.5?
On real professional work Opus 5.5 is 1st of 27 and Muse Wonder 4th, outside the test's margin. On regulatory research the test can't separate them: Opus gets 65.15% at $22.21 a question, Muse Wonder 63.64% at $3.92.
Can Muse Wonder read supplier documents?
No document test has scored it yet. TeleOCR, a free model, leads the independent document test, so compare both on your own TDSs, SDSs, and CoAs.
Can I run Muse Wonder on my own servers?
No. Its weights are closed, so it runs only on Meta's service, through the Meta Model API or Muse Code.
Sources.
- Artificial Analysis: models and the Intelligence Index Artificial Analysis, an independent testing company. Each model's record gives its index score, its cost per index task, its price, and its professional work rating.States no date · Vailent checked 12 Oct 2026
- Artificial Analysis: GDPval-AA Artificial Analysis, an independent testing company. It rates models on real professional tasks and publishes a 95% interval for each.States no date · Vailent checked 12 Oct 2026
- Artificial Analysis: Muse Wonder Artificial Analysis, an independent testing company. Its record gives the release date, the context window, and whether the weights are open.States no date · Vailent checked 12 Oct 2026
- vals.ai Legal Research, regulatory questions vals.ai, an independent testing company. Lawyers wrote and checked the questions; an answer counts only if it meets every point on the lawyers' list.Published 12 Oct 2026 · Vailent checked 12 Oct 2026
- OmniDocBench leaderboard OpenDataLab (Shanghai AI Lab), which also makes MinerU, a free model it ranks.Published 11 Sep 2026 · Vailent checked 12 Oct 2026
- Meta AI Research: Muse Wonder The maker's own page.Published 12 Oct 2026 · Vailent checked 12 Oct 2026
- Meta: Introducing Muse Wonder and Muse Wonder Full The maker's own page.Published 12 Oct 2026 · Vailent checked 12 Oct 2026