Meta launched Muse Wonder and Muse Wonder Full on October 12, 2026, presenting them as agents built to handle work across conversations and tasks. Developers can access Muse Wonder through Meta's API. The higher-scoring Wonder Full is limited to Muse Code subscribers and is not available through the API.

That access gap matters for developers comparing AI providers. They can integrate one model directly, but cannot test the more capable version through the same route.

What developers can access

TechCrunch reported these API prices for Muse Wonder:

  • $4 per million input tokens
  • $20 per million output tokens

TechCrunch also reported that Wonder Full is available only to Muse Code subscribers, not through the API.

Meta says both models use Active Memory to carry relevant details across conversations. It says Active Focus helps them work through obstacles, change approaches and ask for clarification. Meta plans to bring the models to its Muse assistant, first for subscribers.

What the independent scores show

Artificial Analysis scored both models on its Intelligence Index using maximum reasoning effort:

  • Muse Wonder Full: 56
  • Muse Wonder: 52
  • Meta's earlier Muse Spark 1.3: 48

Wonder Full placed in the top three on six of the index's 10 evaluations, according to Artificial Analysis. The evaluator identified knowledge questions as a relative weakness. The results show progress on specific tests, not that the model leads across every task.

Meta's research post makes broader performance claims and describes tests of memory, coding and long-running tasks. Those results come from the company's own evaluation suite. Meta said a fuller safety evaluation report would follow, so part of the public evidence is still pending.

The tradeoff for developers

For U.S. developers, the practical distinction is clear: the model with the higher independent score cannot be integrated directly through Meta's API. They can test Muse Wonder through the API, but assessing Wonder Full requires a Muse Code subscription.

Meta's release pairs API access to Muse Wonder with subscriber-only access to Wonder Full. Independent benchmarks offer a promising, but limited, comparison while Meta's fuller safety report remains pending.

The launch marks a step in Meta's effort to challenge OpenAI and Anthropic. How quickly outside developers can judge Wonder Full will depend in part on whether Meta expands access beyond its coding subscription.