Meta Muse Wonder puts Meta back near the top of the independent rankings, but its strongest version isn’t in the API. Meta released Muse Wonder and Muse Wonder Full on 12 October 2026. Wonder costs $4 per million input tokens and $20 per million output tokens. Wonder Full is limited to paid Muse Code subscribers. Here is the price, the scores and who can use what.
Key takeaways
- Two models: Muse Wonder (API and Muse Code) and Muse Wonder Full (Muse Code subscribers only).
- Muse Wonder API price: $4 input and $20 output per million tokens, cached input $0.20, per VentureBeat. Same list price as Claude Opus 5.5.
- Scores conflict: VentureBeat reports Artificial Analysis scores of 56 (Full) and 52 (Wonder). Artificial Analysis’s own page now shows 50 for Full.
- Closed weights: neither model can be downloaded or self-hosted.
- Privacy: Muse Code subscriptions default to a mode that lets Meta train on your prompts and code. The pay-as-you-go API doesn’t.
Last updated: 1 November 2026 · Status: live · I’ll update this post if Meta opens Muse Wonder Full to the API.
[Insert “Table of Contents by Rank Math” block here]
What is Meta Muse Wonder?
Meta describes Muse Wonder and Muse Wonder Full as its most capable models to date, built for personal and professional agentic tasks, from navigating an insurance claim to building a working app. They follow the earlier Muse Spark family. Meta says they bring skills in law, finance and health, plus frontier-level coding.
Two features are central. Active Memory carries what matters across conversations. Active Focus keeps an agent working when a task gets hard: it tries other approaches, recovers from errors and checks in before anything important. Both models accept text, images, audio and video, output text, and have a context window of about one million tokens, according to VentureBeat.
Who can use Muse Wonder Full and who can use Wonder
| Model | Where | Who |
|---|---|---|
| Muse Wonder | Meta Model API (muse-wonder-1.0) and Muse Code | Developers and Muse Code users |
| Muse Wonder Full | Muse Code only | Subscribers: Power $50/month, Maximum $200/month |
| Neither | Open weights | Not downloadable or self-hostable |
This split is the main buyer’s problem. VentureBeat notes that the stronger Wonder Full has no pay-as-you-go price and can’t be built into your own product through the API. If you want to ship a feature, you can only use the weaker model today.
Muse Wonder API price compared
| Model | Input / 1M | Output / 1M | Cached input / 1M |
|---|---|---|---|
| Meta Muse Spark 1.3 | $1.25 | $4.25 | $0.15 |
| Meta Muse Wonder | $4.00 | $20.00 | $0.20 |
| Claude Opus 5.5 | $4.00 | $20.00 | $0.20 |
| GPT-6 Astra | $10.00 | $50.00 | $1.00 |
| Claude Fable 5.1 | $10.00 | $50.00 | $0.20 |
Web search through the API adds $2.50 per 1,000 queries. Meta lists limits of 3,000 requests and four million tokens per minute. As a worked example, an agent run that reads 200,000 input tokens and writes 20,000 output tokens costs about $1.20 at list price ($0.80 input plus $0.40 output).
A low token price doesn’t guarantee a low bill. Artificial Analysis estimates its average Intelligence Index task costs $4.52 on Wonder, against $5.98 on Claude Opus 5.5 and $3.26 on GPT-6 Astra, because models use different numbers of tokens.
Meta Muse Wonder benchmarks: what is verified
Here is where it gets messy. Meta Muse Wonder benchmarks come from two kinds of source, and they disagree.
| Source | Muse Wonder Full | Muse Wonder | Claude Opus 5.5 | GPT-6 Astra |
|---|---|---|---|---|
| Artificial Analysis Index at launch (via VentureBeat) | 56 | 52 | 58 | 53 |
| Artificial Analysis page, 1 Nov 2026 (Index v5.0) | 50 | Not checked | Not checked | Not checked |
On launch day, VentureBeat reported Full at 56, tying Claude Sonnet 5.5 and ahead of GPT-6 Astra, Gemini 4 Argon and Claude Fable 5.1 at 53, behind only Opus 5.5 at 58. Artificial Analysis’s own page now lists Full at 50 on its updated Intelligence Index v5.0, and says there’s no publicly available API.
Meta’s model card adds task-level numbers. Meta says Wonder Full scores 57.5 on SWE-Marathon (long coding tasks) against 46.3 for Opus 5.5 and 42.5 for GPT-6 Astra, and 13.1 on Harvey’s legal benchmark against 4.2 for Opus 5.5. It trails Opus 5.5 and GPT-6 Astra on Terminal-Bench 4.0 and HealthBench Pro. VentureBeat points out that the card mixes third-party and Meta-run tests, and some use different coding tools.
Privacy: the Muse Wonder Full catch
According to VentureBeat, Meta offers a Standard mode, where submitted content isn’t used for training, and a Contributor mode, where it is. The pay-as-you-go API runs only in Standard mode. Muse Code subscriptions, which are the only way to reach Muse Wonder Full, default to Contributor mode. You can switch it off, but only for future submissions. Don’t put confidential code or customer data into a Contributor-mode account.
What it means for India
For Indian developers, the Muse Wonder API price is the easy part: multiply the dollar rates by today’s USD-INR rate and compare with Claude Opus 5.5, which costs the same per token. The harder question is whether you need Wonder at all. If your work is long coding tasks or document-heavy agents, test it on your own jobs. For simple chat or classification, cheaper models cost far less.
Watch: Meta Muse Wonder, explained
[Paste your YouTube or Instagram video link on an empty line below.]
What we don’t know yet
- Whether Muse Wonder Full will ever get a pay-as-you-go API price.
- Why Artificial Analysis shows 50 for Wonder Full today when launch coverage reported 56.
- How Wonder performs on your own tasks. Meta’s model card mixes independent and in-house tests.
- When Wonder arrives inside the Muse app, which Meta says will reach subscribers first.
FAQ
How much does Meta Muse Wonder cost?
Through the Meta Model API it costs $4 per million input tokens and $20 per million output tokens, with cached input at $0.20.
Is Muse Wonder Full available through the API?
No. Muse Wonder Full is limited to Muse Code subscribers on the $50 and $200 monthly plans.
Is Muse Wonder open source?
No. Neither Wonder model has downloadable weights.
Is Muse Wonder better than Claude Opus 5.5?
Not overall. Launch coverage had Opus 5.5 ahead on Artificial Analysis’s index. Wonder Full led on some coding and legal tests in Meta’s own card.
Hari’s take
Meta Muse Wonder is a serious model with a confusing release. The best version sits behind a coding subscription, the cheaper one is priced like Opus, and the score depends on which date you check. If you build products, test Wonder against Opus on your own work before you switch anything. If you just want to try Meta’s best, you’ll need a Muse Code plan and a careful look at the privacy setting.
Would you pay for Muse Code? Tell me in the comments.
Sources: Meta Newsroom: Introducing Muse Wonder and Muse Wonder Full, Meta: Muse Wonder, VentureBeat, Artificial Analysis: Muse Wonder Full, CNBC, Online Tech Tips. Last updated 1 November 2026.
