Muse Wonder · Meta

What does Muse Wonder cost to run in a coding agent?

Muse Wonder costs about $0.11 per average coding-agent request on our measured traffic, at list prices of $4 per million input and $20 per million output tokens. It scores 52 on the Artificial Analysis Intelligence Index, 6th of the 48 models we track. Source: Artificial Analysis (artificialanalysis.ai)

List prices: OpenRouter (vendor endpoint), checked 2026-10-13. Request costs: Standard Compute gateway data, 79,301 coding-agent requests from 75 users, Sep 13 to Oct 12, 2026. How we compute this

Run current Claude, GPT, Gemini and open models in your agent from $19/month

See plans

What Muse Wonder costs per coding-agent request

Cost at list price	Muse Wonder
Prompt under 100K tokens50% of requests, 87% of input cached	$0.052
Prompt of 100K to 272K39% of requests, 88% of input cached	$0.13
Prompt over 272K tokens11% of requests, 86% of input cached	$0.28
Average request	$0.11
100 requests	$10.55
Median session16 requests; a session ends after 10 idle minutes	$1.43
A month of regular usemedian of 20 users active on 5+ days	$152

Each logged request priced at the vendor’s list rates for its own token counts, including cache reads and any higher rate for long prompts. Average output: 698 tokens per request.

Muse Wonder benchmarks, speed and latency

Artificial Analysis	Value	Rank
Intelligence Index10 evaluations, 0 to 100	52	6 of 48
Terminal-Bench 4.0agentic coding in a terminal	52.0%	6 of 39
Output speedmedian	115 tokens/s	15 of 44
Time to first answer tokenincludes thinking time	38 s	20 of 44

Source: Artificial Analysis (artificialanalysis.ai). Measured variant: Muse Wonder (Max), speed on Meta. Fetched October 13, 2026. Rank is among the models on this site, older versions included. Artificial Analysis runs Terminal-Bench 4.0 in its own setup, so its scores differ from the tbench.ai leaderboard.

About Muse Wonder

Muse Wonder is a frontier reasoning model from Meta, on OpenRouter since October 12, 2026 and served there by Meta itself. Meta designed it for complex personal and professional agent tasks; it reads text, images, video, audio and PDFs in a 1M-token window. Unlike Llama, the weights are not open. Artificial Analysis scores it 52 on its Intelligence Index (max effort) and 52.0% in its Terminal-Bench 4.0 run, at about 115 output tokens per second. Its $4/$20 price matches Claude Opus 5.5, so it costs about the same per request in our measurements ($0.106 vs $0.107) while scoring below Opus 5.5 on both boards.

Sources: OpenRouter listing · Artificial Analysis

Muse Wonder FAQ

How much does Muse Wonder cost per coding-agent request?

$0.11 on average at list price, $10.55 per 100 requests, and $1.43 for a median 16-request session. That is 79,301 real coding-agent requests (Sep 13 to Oct 12, 2026) priced at Meta's list rates; the median request sent 99,534 input tokens, 87% of input was read from cache, and requests returned 698 output tokens on average.

Does Muse Wonder charge more for long prompts?

No. Muse Wonder bills the same rate per token up to its 1M-token context window.

How fast is Muse Wonder?

Artificial Analysis measures Muse Wonder (Max) at 115 output tokens per second, with 38 seconds to the first answer token, thinking time included, on Meta's API. Source: Artificial Analysis (artificialanalysis.ai).

Is Muse Wonder good for AI agents?

Multimodal agent work where you want a frontier model outside the usual three vendors. On our scoring, which is calibrated to public benchmarks, it rates 9.0/10 for agentic ability and 9.3/10 for output quality. 52 on the Artificial Analysis Intelligence Index, in the frontier group.

What is Muse Wonder's context window?

Muse Wonder has a 1.0M-token context window, large enough to hold sizeable codebases or document sets in a single run. Meta is the maker.

Can I self-host Muse Wonder?

No. Muse Wonder is a proprietary, hosted-only model from Meta. You access it through an API, such as OpenRouter, rather than running the weights yourself.

What are the downsides of Muse Wonder?

The main trade-offs: Same price as Claude Opus 5.5 ($4/$20) but lower on Artificial Analysis and Terminal-Bench 4.0; Closed weights, unlike Meta's Llama models; Released October 2026, so its track record in coding agents is short. It's strongest for multimodal agent work where you want a frontier model outside the usual three vendors.