Releases

Muse Wonder 1.0

MiniRouter and Muse Wonder 1.0 by Meta, separated by a silver fold of light on black.

Meta’s new model for complex coding and agents that work through long tasks.

Muse Wonder 1.0 is available on MiniRouter as . The model page has current pricing, provider routes and API setup.

View Muse Wonder 1.0

At a glance

Context
1,048,576 tokens
Publisher context window.
Best fit
Coding and agents
Built for multi-file engineering and sustained tool use.
Meta API inputs
Multimodal → text
Meta documents text, image, video, audio and PDF input.
Meta API reasoning
Minimal through max
Meta’s API requires reasoning; none is unsupported.

Publisher capabilities describe Meta’s API. Check MiniRouter’s supported inputs and parameters.

Coding and computer use

Two evaluations from Meta’s launch report. Each compares models within a shared setup; neither measures MiniRouter performance.

ProgramBench

Build a program that passes at least 95% of its hidden tests.

  • Wonder50.0%
  • Full*59.0%
  • Fable 5.155.5%
  • Astra52.5%
  • Opus 5.570.5%

0 to 100% · higher is better

Share of successful rollouts. All five models use mini-swe-agent at max effort, with internet disabled after setup.

*Full is a separate Muse Code subscription model.

Meta-reported, 12 October 2026. Results · Evaluation methods
View chart data
ProgramBench comparison data
ModelSuccess rate
Muse Wonder50.0%
Muse Wonder Full59.0%
Claude Fable 5.155.5%
GPT-6 Astra52.5%
Claude Opus 5.570.5%

OSWorld V2.1

Use a computer to complete desktop workflows.

  • Wonder72.6
  • Full*78.4
  • Fable 5.180.7
  • Astra75.8

0 to 100 points · higher is better

Mean partial reward across 108 workflows, not a task-completion percentage. These four models share Meta’s evaluation framework at max effort. Opus is omitted because Meta cites Anthropic’s reported result.

*Full is a separate Muse Code subscription model.

Meta-reported, 12 October 2026. Results · Evaluation methods
View chart data
OSWorld V2.1 comparison data
ModelPartial reward
Muse Wonder72.6 points
Muse Wonder Full78.4 points
Claude Fable 5.180.7 points
GPT-6 Astra75.8 points

Computer use requires an agent and tools in your application. Selecting the model does not supply that integration.

Publisher pricing

Meta Standard rates per million tokens, before MiniRouter markup.

Input
$4.00
Cached input
$0.20
Output
$20.00

No long-context premium. See current MiniRouter rates →

Budget for thinking and the answer

Your input

Output token limit

ReasoningFinal answer
Both consume output tokens and use the output rate. The split varies by request.

More reasoning effort does not select Wonder Full. Full is available separately through selected Muse Code subscriptions.

Try it

Start with a text request and leave room in the output budget for reasoning.

Muse Wonder 1.0 · cURL
curl https://api.minirouter.sh/v1/chat/completions \
  -H "Authorization: Bearer $MINIROUTER_KEY" \
  -H 'Content-Type: application/json' \
  --data '{
    "model": "meta/muse-wonder-1.0",
    "reasoning_effort": "high",
    "max_tokens": 8192,
    "messages": [{ "role": "user", "content": "Write a Python CSV parser and tests for quoted fields." }]
  }'

More languages, client setup and supported effort settings →

Keep exploring

Compare published results in the model rankings and coding benchmarks, or read more model releases.

For MiniRouter product updates, visit the changelog.

Sources

  1. Meta Muse Wonder announcement

    Published Oct 12, 2026

    Read at Meta ↗
  2. Meta model documentation

    Published Oct 12, 2026

    Read at Meta ↗
  3. Meta Model API pricing

    Published Oct 12, 2026

    Read at Meta ↗
  4. Meta reasoning guide

    Published Oct 12, 2026

    Read at Meta ↗
  5. Muse Wonder benchmark results

    Published Oct 12, 2026

    Read at Meta ↗
  6. Muse Wonder evaluation methods

    Published Oct 12, 2026

    Read at Meta ↗

Editorial reference revised 2026-10-12. Catalog status comes from the generated routing snapshot.

Every model, one key.