Meta's Muse Wonder Ships Closed at $4 and $20 Per Million Tokens, Ranked Ninth by Artificial Analysis
A.I. / news
Meta's Muse Wonder Ships Closed at $4 and $20 Per Million Tokens, Ranked Ninth by Artificial Analysis
Meta's October 12 post gives no benchmark, price or weights policy. Artificial Analysis supplies them, along with a 39-second wait for the first token.
Meta released Muse Wonder on October 12 as a closed model priced at $4 per million input tokens and $20 per million output tokens, and Artificial Analysis ranks it ninth of 216 models on its Intelligence Index.
Meta announced the Muse Wonder series in a newsroom post dated October 12 and updated October 13. Muse Wonder is available now through the Meta Model API and Muse Code. A larger variant, Muse Wonder Full, is limited to Muse Code subscribers, and the post mentions no API access for it.
What Meta's October 12 post left out
The post calls both models Meta's "most capable models to date" and says they build on Muse Spark. It names two features. Active Memory carries context across conversations, and Active Focus keeps working toward a goal when tasks get hard and checks in before important actions.
It names no benchmark, no price, no context window and no parameter count. It does not say whether the weights are open. It defers evaluation detail to a research post and says "a full report with additional safety evaluations will be available soon," with no date.
The name of the larger model is also in dispute. Euronews, reporting with AFP on October 13, calls it "Wonderful." Meta's post calls it Muse Wonder Full.
What Artificial Analysis shows for Muse Wonder
Artificial Analysis lists Muse Wonder (Max) as proprietary, with a 1-million-token context window. It accepts text, image, speech and video and outputs text only.
| Measure | Muse Wonder | Artificial Analysis median |
|---|---|---|
| Intelligence Index | 52 (rank 9 of 216) | 27 |
| Input price per 1M tokens | $4.00 | $2.00 |
| Output price per 1M tokens | $20.00 | $10.00 |
| Output speed, tokens per second | about 131 | 84 |
A full run of the Intelligence Index costs about $4.52 on Muse Wonder, per the same page. The model generated 130 million tokens on the index against a median of 85 million.
- Muse Wonder (Max)39 s
- Median, comparable models5.87 s
Source: Artificial Analysis model page for Muse Wonder (Max), accessed 2026-10-14
Time to first token is about 39 seconds, at the high end for its price tier. The page lists two API providers.
Meta's shift from Llama to Muse
Euronews says Meta cited Artificial Analysis data, though the post text carries no such figure, and reports that Muse Spark launched in April as a break from the open-source Llama line. Mark Zuckerberg, Meta's chief executive, said on X that Muse Wonder "is a much larger and more advanced class of models that brings us to the frontier." He also said Meta "delayed shipping Muse for several months to focus on safety and security." Euronews attributes both quotes to him.
Alexandr Wang, whom Euronews says oversees Meta Superintelligence Labs, called Wonder "our largest and most capable model yet."
Meta's open-weight work has not stopped. DataNorth's Qwen3.8-27B review says Meta released Muse Glimmer-30B under Apache 2.0 on August 10. Muse Wonder is the opposite: closed weights, a metered API and a subscription tier.
Euronews says Muse Wonder could cost $20 to $100 a month in a subscription. Meta's post gives no subscription price, so that figure is Euronews's alone. Compare Google's Gemini 4 Argon pricing, which is $2 and $10 per million tokens until November 30, and its safety write-up for how another lab documented a release.
Meta said the full safety report will follow, and the post says Muse Wonder will "soon power more of Muse," beginning with subscribers.
Sources
More in A.I.
- 01Torvalds Posts a Kernel Patch Series Written With Claude Code, and Sources Disagree on Its Size and SpeedupLWN counts ten patches and an order-of-magnitude gain, while It's FOSS counts eight patches and a threefold gain on one laptop.
- 02Parsimony Ranks Coding Agents by How Much Code Their SWE-bench Fixes Add, and Its Own Page Warns Smaller Is Not BetterThe MIT-licensed benchmark counts coding units in published patches without rerunning a single test, and reports that code volume tracks solve rate.
- 03OpenBMB's MiniCPM-V 4.7 Went Up With No Model Card on Oct. 6, and Now Reads Apache-2.0 With No Benchmark NumbersThe 35-billion-parameter mixture-of-experts vision model is 27 times larger than its predecessor, and its card claims a win over Qwen3.6 35B-A3B without a score.
- 04Design Creativity Bench Scores GPT-6 Astra 65.5 Against Humans' 92.3, and the Paper Never Names Its JudgeA post using Simon Willison's pelican SVGs to argue that models converge on one image cites a leaderboard whose method the arXiv page does not describe.