Meta Launches Muse Wonder AI Models With Improved Memory and Coding

Meta has released Muse Wonder and Muse Wonder Full, two new AI models designed to remember context across conversations and work through complex coding and other multi-step tasks.

Wonder is available today through the Meta Model API and Muse Code, while Wonder Full is limited to the coding service's Power Usage and Maximum Usage subscription plans. Both models support a one-million-token context window.

A feature called Active Memory carries important information across conversations and retrieves it when it becomes relevant. In Meta's internal MemoryBench evaluation, Wonder scored 85.6% and Wonder Full scored 85.5% at maximum reasoning effort, compared with 81.5% for Claude Opus 5.5 and 81.4% for GPT-6 Astra. The test measures how well models remember and use context over a long interaction history.

Active Focus is designed to help the models keep working when a task gets difficult, try alternative approaches, and recover from errors. Along the way, they can ask for clarification and check with the user before important actions. The models are also trained to avoid sharing private information without permission.

For Mac users, the new models are accessible through Muse Code, Meta's terminal-based coding agent. They can plan longer projects, delegate work to sub-agents, and make and verify changes across multiple files.

The published SWE-Marathon results, which test long-running software engineering work, put Wonder Full at 57.5% using its xhigh reasoning setting and Wonder at 53.1% using maximum reasoning. Claude Opus 5.5 scored 46.3% and GPT-6 Astra scored 42.5% in the comparison.

Wonder will reach the Muse personal AI agent later, starting with subscribers, and will also power artifact generation and other tasks that need more capable reasoning. The service, which recently expanded to iPad, handles jobs such as email, shopping, and trip planning. A date for the model upgrade has not been announced.

API pricing for Wonder is $4 per million input tokens, $0.20 per million cached input tokens, and $20 per million output tokens. Wonder Full is included with Muse Code's $50-per-month Power Usage and $200-per-month Maximum Usage plans, rather than being sold through the API.