# Meta's Muse Wonder Full Ties for Second on Artificial Analysis, but You Can Only Get It Through a Muse Code Subscription

Meta released two frontier models on 12 October. The cheaper one is on the API at $4 and $20 per million tokens. The stronger one is only inside Muse Code, where the default setting lets Meta train on your code.

By Priya Nair (AI Industry Reporter) — published 2026-10-14, updated 2026-10-14
Source: https://aiscoutdaily.com/news/meta-muse-wonder-full-not-on-api-price-scores

## TL;DR
Meta Superintelligence Labs released Muse Wonder and Muse Wonder Full on 12 October 2026. On Artificial Analysis's Intelligence Index, Wonder Full scores 56, tied with Claude Sonnet 5.5 and behind only Claude Opus 5.5 (58), while Muse Wonder scores 52. Muse Wonder is on the Meta Model API at $4 per million input tokens and $20 per million output tokens with a 1M-token context window; Wonder Full has no per-token price and is sold only through Muse Code's $50 and $200 monthly plans. Muse Code subscriptions default to a Contributor mode that allows Meta to train on prompts and code, unlike the API.

Meta's new models arrived six weeks after Muse Spark 1.3 and are a real step up in independent testing. But the release splits in two: one model you can call from your own code, and a stronger one you can only use inside Meta's coding product.

## What Meta released

- Muse Wonder (model ID muse-wonder-1.0): available now on the Meta Model API and in Muse Code. 1M-token context window.
- Muse Wonder Full (muse-wonder-full-1.0): available only to Muse Code subscribers on the Power Usage and Maximum Usage plans. No per-token price. 1M-token context window.
- Both: no downloadable weights. Meta says Muse Wonder will "soon" power its Muse app, rolling out first to subscribers.
- Features Meta highlights: Active Memory, which carries context about you across conversations, and Active Focus, which keeps working toward a goal by trying other approaches and checking in before important actions.

## How they score, measured by someone other than Meta

| Model | Intelligence Index v4.3 | Cost per Index task | API price per 1M tokens (in / out) |
| --- | --- | --- | --- |
| Claude Opus 5.5 | 58 | $5.98 | $4 / not stated |
| Muse Wonder Full | 56 | n/a | Not on the API |
| Claude Sonnet 5.5 | 56 | $5.46 | $2 / $10 |
| GPT-6 Astra | 53 | $3.26 | $10 / $50 |
| Gemini 4 Argon (high) | 53 | $1.99 | $2 / $10 |
| Claude Fable 5.1 | 53 | $7.63 | $10 / $50 |
| Muse Wonder | 52 | $4.52 | $4 / $20 |
| Muse Spark 1.3 | 48 | $1.60 | $1.25 / $4.25 |

*Artificial Analysis, 12 October 2026. Wonder Full has no per-token price, so no cost figure. Opus 5.5 output price not given in the source.*

Two details stand out in Artificial Analysis's numbers. On its Coding Agents Index, Wonder Full inside Muse Code scores 68, level with Claude Code running Sonnet 5.5 and above Claude Code with Opus 5.5 (66). And on AA-Omniscience, a knowledge and hallucination test, Wonder Full scores 30 against 46 for Opus 5.5 and 42 for Gemini 4 Argon, so it is not ahead everywhere.

Meta's own table, reported by VentureBeat, shows Wonder Full ahead of Opus 5.5 on SWE-Marathon (57.5 vs 46.3) and Harvey's legal benchmark, while Opus 5.5 leads on HealthBench Pro and Opus 5.5 and GPT-6 Astra lead on Terminal-Bench 4.0. Meta's blog post itself gives no numeric scores in text, and its full safety report is "available soon".

## The price is per token, the bill is per task

Muse Wonder's $4 input price matches Claude Opus 5.5, but it uses a lot of input: about 7.7 million input tokens per Intelligence Index task, against about 6 million for Muse Spark 1.3. That is why it costs $4.52 per task to run, 2.8 times Muse Spark 1.3, and more than GPT-6 Astra ($3.26) despite Astra's higher token prices. Caching matters: of that $4.52, $3.43 is input, and most of it is cache writes and hits.

## The data setting most coverage skips

- API: falls under Meta's Standard mode, where Meta says submitted content is not used for training. API limits are 3,000 requests and 4 million tokens a minute; web search costs $2.50 per 1,000 queries.
- Muse Code subscriptions (the only way to use Wonder Full): default to Contributor mode, which allows training on prompts, code, files, and outputs. Turning it off applies only to future submissions, and Meta's terms say not to send confidential or personal information in this mode.
- Zero data retention: available to approved API organisations, but it excludes server-side conversation state, background requests, file uploads, and built-in web search.

## What this means if you're choosing a model

If you build on an API, Muse Wonder is a credible new option at Opus-level token prices, but test it on your own tasks, because its token appetite can make it cost more per job than cheaper-looking rivals. If you want the stronger Wonder Full, you are choosing Muse Code as a product, at $50 or $200 a month, and you should switch off Contributor mode before you open a private repository in it.

## Sources
- [Meta: Muse Wonder (Oct 12, 2026)](https://research.meta.ai/blog/muse-wonder)
- [Meta developer docs: Muse Wonder model page](https://dev.meta.ai/models/muse-wonder)
- [Artificial Analysis: Muse Wonder and Muse Wonder Full (Oct 12, 2026)](https://artificialanalysis.ai/articles/muse-wonder)
- [VentureBeat: Meta's new Muse Wonder and Muse Wonder Full models achieve frontier performance, but only the weaker one is available via API](https://venturebeat.com/technology/metas-new-muse-wonder-and-muse-wonder-full-models-achieve-frontier-performance-but-only-the-weaker-one-is-available-via-api)
- [Electronic Times: Meta unveils Muse Wonder series (Oct 14, 2026)](https://en.etnews.com/20261014200002)