Muse Wonder
- Provider
- Meta
- Status
- Current
- Context
- 1,000,000 tok
- Price
- $4 / $20 /MTok
Muse Wonder is Meta’s frontier model on the Meta Model API, released on 12 October 2026 alongside the stronger Muse Wonder Full (Meta). It costs $4/$20 per million tokens, the same token price as Claude Opus 5.5, and is also available in Muse Code, Meta’s coding app.
Artificial Analysis scored it 52 on the Intelligence Index v4.3.2 at max effort, level with GPT-6.1 Sol, at $4.52 per completed index task. Meta published no SWE-bench figure, no parameter count and no knowledge cutoff.
Quick specs
| Muse Wonder | |
|---|---|
| Released | 12 October 2026 |
| Input price | $4.00 / MTok |
| Cached input | $0.20 / MTok |
| Output price | $20.00 / MTok |
| Web search | $2.50 per 1,000 queries |
| Context window | About 1,000,000 tokens |
| Max output | Not published |
| Knowledge cutoff | Not published |
| AA Intelligence Index | 52 (max) |
| Cost per AA index task | $4.52 |
| Output speed | 87.9 tokens a second (Artificial Analysis) |
| API model ID | muse-wonder-1.0 |
| Rate limits | 3,000 requests and 4 million tokens a minute |
Prices, model ID and rate limits are from VentureBeat, citing Meta’s documentation. Artificial Analysis lists the same $4/$20.
Muse Wonder and Muse Wonder Full
Meta released two models under the Wonder name. Muse Wonder is the one developers can buy by the token. Muse Wonder Full is available only to Muse Code subscribers and has no public API.
| Muse Wonder | Muse Wonder Full | |
|---|---|---|
| AA Intelligence Index (max) | 52 | 56 |
| Meta Model API | Yes, $4 / $20 | No |
| Muse Code | Yes | Subscribers only |
| Cost per AA index task | $4.52 | Not measurable without an API |
Meta has not said how the two differ technically. Its launch post says both “build on larger, stronger base models” than Muse Spark 1.3, and that both improve as more test-time compute is applied. Artificial Analysis lists them as separate models and measured both at max effort.
What Meta says is new
Active Memory. Meta describes it as “carrying important context about you across conversations and actively retrieving the details relevant to the task at hand”.
Active Focus. Meta’s term for how the model “continues working toward a challenging goal: trying alternative approaches, working through blockers, and adapting to new information”.
Safety. Meta calls Muse Wonder its “most aligned model to date”, with its “strongest safeguards to date against CBRN and cyber misuse”. It says a full report with additional safety evaluations “will be available soon”; that report was not published at launch.
Mark Zuckerberg described Wonder as “a much larger and more advanced class of models” than the Muse Spark family, according to VentureBeat.
Benchmark performance
Independent, from Artificial Analysis:
| Measure | Muse Wonder | Muse Wonder Full | Claude Opus 5.5 | GPT-6 Astra |
|---|---|---|---|---|
| Intelligence Index v4.3.2 (max) | 52 | 56 | 58 | 53 |
| Cost per index task | $4.52 | n/a | $5.98 | $3.26 |
| Output tokens to run the index | 130 million | 140 million | n/a | n/a |
| Price (in/out per MTok) | $4 / $20 | No API | $4 / $20 | $10 / $50 |
Artificial Analysis ranks Muse Wonder 9th of 216 models on its page. It used 130 million output tokens to run the index against a median of 85 million.
Published by Meta, as reported by VentureBeat, which notes the table mixes internal and third-party evaluations run under varying conditions:
| Benchmark | Muse Wonder | Muse Wonder Full | GPT-6 Astra | Claude Opus 5.5 |
|---|---|---|---|---|
| DeepSWE 1.1 (software engineering) | 78.8 | 79.0 | 74.1 | 74.2 |
| SWE-Marathon (long-running coding) | 53.1 | 57.5 | 42.5 | 46.3 |
| Harvey LAB-AA (legal) | 10.0 | 13.1 | 8.6 | 4.2 |
| AutomationBench-AA (workflows) | 39.1 | 45.4 | 41.6 | 41.2 |
| JobBench (professional) | 70.5 | 75.3 | 66.1 | 75.4 |
| Terminal-Bench 4.0-AA | 52.0 | 57.1 | 59.1 | 59.6 |
| HealthBench Pro | 60.5 | 62.3 | 65.3 | 65.6 |
On Meta’s own table Muse Wonder leads GPT-6 Astra and Claude Opus 5.5 on DeepSWE 1.1, SWE-Marathon and the Harvey legal benchmark, and trails both on Terminal-Bench 4.0, AutomationBench and HealthBench Pro.
How it compares
vs GPT-6.1 Sol
GPT-6.1 Sol scores the same 52 on the Artificial Analysis index at $2/$10 per million tokens and $0.72 per completed task. Muse Wonder costs twice as much per token and about six times as much per task.
vs Claude Opus 5.5
Claude Opus 5.5 has the same $4/$20 token price and scores six points higher on the index (58), at $5.98 per completed task against Muse Wonder’s $4.52.
vs Muse Spark 1.3
Muse Spark 1.3 costs $1.25/$4.25 per million tokens. Muse Wonder is Meta’s larger model at about three times the input price and nearly five times the output price.
Data use
The Meta Model API runs in Standard mode, in which submitted content is not used for model training. Muse Code subscriptions default to Contributor mode, which lets Meta use prompts, code, files and outputs to improve future models; subscribers can switch it off for future submissions (VentureBeat).
Known limitations
No SWE-bench figure. Meta published no SWE-bench Pro or Verified result.
Specs withheld. Parameter count, knowledge cutoff and maximum output length are not published.
No open weights. Neither Wonder model has downloadable weights, unlike Meta’s Muse Glimmer.
Cost per task. At $4.52 per completed index task it costs more to run than GPT-6 Astra ($3.26), which scores one point higher on the same index.
Safety report pending. Meta’s full safety report was not available at launch.
Frequently asked questions
When was Muse Wonder released?
12 October 2026, on the Meta Model API and in Muse Code. Meta says it will power the Muse app soon, starting with subscribers.
How much does Muse Wonder cost?
$4 per million input tokens and $20 per million output tokens on the Meta Model API, with cached input at $0.20 per million and web search at $2.50 per 1,000 queries.
What is the difference between Muse Wonder and Muse Wonder Full?
Muse Wonder is sold by the token on the Meta Model API and scores 52 on the Artificial Analysis Intelligence Index. Muse Wonder Full scores 56 and is available only to Muse Code subscribers, with no public API. Meta has not published how the two differ technically.
Is Muse Wonder open source?
No. Muse Wonder and Muse Wonder Full are proprietary, with no downloadable weights.
Is Muse Wonder better than Claude Opus 5.5?
Not on the independent index: Artificial Analysis scores Muse Wonder 52 and Claude Opus 5.5 58, at the same $4/$20 token price. On Meta’s own published table Muse Wonder leads Opus 5.5 on DeepSWE 1.1 (78.8 to 74.2) and SWE-Marathon (53.1 to 46.3) and trails it on Terminal-Bench 4.0 (52.0 to 59.6).