Company Meta✴ released two new AI models – Muse Wonder and Muse Wonder Full, which are positioned as direct competitors to the best models from OpenAI and Anthropic. Access to new products is already open to subscribers of the Muse Code service, and the Wonder model has also become available through Meta✴ API Models.
Image source: Meta✴
The developers claim that the new AI is not inferior in its capabilities to GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.1. Such statements are based on internal testing results in industry benchmarks. At the same time, comparison with Google’s Gemini 4 Argon remains ambiguous, since the capabilities of the model have not yet been fully explored by the market. Most comparisons used the models’ maximum level of reasoning.
In the Personal Superintelligence Index, which combines seven tests of professional and daily tasks, Muse Wonder Full scored 59.1. For comparison, the GPT-6 Astra scored 54.5 points, Claude Opus 5.5 scored 55.2, and Claude Fable 5.1 scored 49.9. The regular Muse Wonder received a score of 54.3.
A noticeable gap from competitors was recorded in programming tasks. In DeepSWE 1.1, which evaluates the ability of AI to solve complex software development problems in existing repositories, Muse Wonder Full scored 79 points against 74.1 for GPT-6 Astra and 74.2 for Claude Opus 5.5. In the SWE-Marathon test, designed for long-term work on large software projects, the result of the older model reached 57.5 points. This is significantly more than 42.5 points for GPT-6 Astra, 45.6 for Claude Fable 5.1 and 46.3 for Claude Opus 5.5.
However, in other tests the advantage of Muse Wonder turned out to be less pronounced. In Terminal-Bench 4.0, which tests execution of tasks on the command line, Muse Wonder Full scored 57.1 points, while GPT-6 Astra and Claude Opus 5.5 received 59.1 and 59.6 points, respectively. In ProgramBench, which evaluates the ability to independently repair a program and pass automatic checks, the result of the Meta model✴ was 59 points – against 70.5 for Claude Opus 5.5.
In the Harvey LAB-AA test, which focuses on completing legal tasks based on source documents and detailed instructions, the Muse Wonder Full was again significantly ahead of the competition: 13.1 points versus 8.6 for the GPT-6 Astra, 6.4 for the Claude Fable 5.1 and 4.2 for the Claude Opus 5.5.
Meta perspective✴ plans to use the Wonder architecture as the basis for its AI agent Muse. First, paid subscribers will have access to the updated agent, then the company intends to develop specialized versions for a wider audience. Specific implementation dates and technical details have not yet been disclosed.
The launch of new models became part of Meta’s aggressive strategy✴ to restore influence in the field of AI. The corporation has invested billions of dollars in talent acquisition and infrastructure development, while expanding its product line, including its Muse Image imaging model.
If you notice an error, select it with the mouse and press CTRL+ENTER.
