Muse Wonder Full Is Meta's New AI Model That Beats GPT-6 Astra in Key Tests

Meta has announced two new AI models, Muse Wonder and Muse Wonder Full, positioning them as competitors to the latest models from OpenAI, Anthropic, and Google. According to the company’s benchmarks, Muse Wonder Full outperforms several leading AI models in coding and legal reasoning tasks.

The announcement follows Meta’s recent launch of Muse, its personal AI agent, which can perform real-world actions on behalf of users. The service quickly gained traction, reaching 700,000 daily users while experiencing stability issues.

Meta developed the new models using its gigawatt-scale Prometheus data center, introducing improvements to pretraining and reinforcement learning.

Muse Wonder introduces Active Memory and Active Focus

Meta has equipped both models with two features designed to improve their ability to handle complex tasks.

Active Memory allows the models to remember important information across conversations and retrieve it when necessary. This helps users maintain context without repeatedly providing the same information.

Active Focus enables the models to work through complicated tasks, explore alternative approaches when something fails, and request clarification when necessary.

These features aim to make Muse Wonder more useful for extended workflows and tasks that require multiple steps.

Muse Wonder Full beats GPT-6 Astra and Claude Opus 5.5 in several benchmarks

According to Meta’s published benchmark results, Muse Wonder Full delivers competitive performance against GPT-6 Astra, Claude Opus 5.5, and Claude Fable 5.1.

The company highlighted several results:

• SWE-Marathon: Muse Wonder Full achieved 57.5%, compared to 46.3% for Claude Opus 5.5 and 42.5% for GPT-6 Astra.
• DeepSWE v1.1: Meta’s model scored 79%, ahead of Claude Opus 5.5 at 74.2% and GPT-6 Astra at 74.1%.
• Harvey LAB-AA: Muse Wonder Full scored 13.1 on legal tasks, compared to 8.6 for GPT-6 Astra and 4.2 for Claude Opus 5.5.
• Artificial Analysis Intelligence Index: Muse Wonder Full reached 56 points, trailing Claude Opus 5.5 at 58 but surpassing GPT-6 Astra and Claude Fable 5.1, which both scored 53.

While Meta’s results suggest strong performance in software engineering and legal tasks, Muse Wonder Full does not lead every benchmark. These comparisons reflect the company’s published results rather than independent testing.

Both models support 1 million tokens, but availability differs

Muse Wonder and Muse Wonder Full support a 1-million-token context window, allowing them to process large documents, lengthy conversations, and extensive codebases.

Both models can understand text, images, audio, and video, although they currently generate text-only responses.

Meta offers Muse Wonder through its Meta Model API and Muse Code, while Muse Wonder Full is available to Muse Code subscribers.

API pricing starts at $4 per million input tokens and $20 per million output tokens.

The launch expands Meta’s AI lineup as the company continues developing its Muse ecosystem. Meta is also preparing to bring Muse to Windows, meaning the personal AI agent will not remain exclusive to mobile devices.