Meta Platforms announced two new versions of its artificial intelligence model through its official announcement – Muse Wonder and Muse Wonder Full. The company says these models are its most capable to date and can compete with OpenAI ChatGPT and Anthropic Claude, especially in solving coding, finance, and legal tasks.
The first model in this series, Muse Wonder, is available through the Meta Model API and the Muse Code platform. Its „Full“ version, Muse Wonder Full, is available to subscribers using the Muse Code service. Both models are designed to create „frontier“ agents who can perform longer, multi-step tasks, such as processing insurance claims or writing software code.
Latest features: Active Memory and Active Focus
The SiliconANGLE source highlights two main innovations – Active Memory and Active Focus. Active Memory allows the model to maintain context about the user across multiple conversation sessions, so important data can be retrieved only when it becomes necessary. Active Focus, on the other hand, helps the model not give up when the task becomes difficult: the model tries alternative strategies, reproduces errors, and checks for user consent before taking important actions.
Benchmark results and competitive comparison
SiliconANGLE provides a detailed comparison with other leaders. On the model card, Muse Wonder Full achieved 13,1 % successful results in the Harvey LAB‑AA legal agent benchmark, while the best competitor, GPT‑6 Astra, achieved 8.6 %. Long-term encoding tasks (DeepSWE 1.1 and SWE‑Marathon) also show that Muse Wonder Full stands out, although some results are not directly controlled. Other benchmarks, such as ProgramBench, HealthBench Pro and OSWorld V2.1, show that Claude Opus 5.5 remains a strong competitor, but Meta emphasizes that their tests were run in „maximum reasoning“ mode.
Availability and pricing
Forkast.news confirms that Muse Wonder is available via the OpenRouter platform for $4 per million input tokens and $20 per million output tokens, with a context box of 1 million tokens. The model also supports multimodal input formats – text, images, video, audio, and PDF documents – as well as tool invocation and structured output formats. Muse Wonder Full is available only to Muse Code subscribers, and its pricing is tied to the Power Usage and Maximum Usage subscription levels.
Security measures and future plans
Meta emphasizes that the models have been trained to follow security guidelines: they should not share private information without permission, refuse malicious requests, and contribute to strengthening cybersecurity. While the company has not yet released a detailed security report, the model card states that Muse Wonder Full can be classified as „moderate or lower“ in three risk domains – chemical, biological, and cyber – when the appropriate security filters are applied.
Market reaction and strategic context
The Business Standard report calls the model a „huge leap“ after Metos„ aggressive AI team building and infrastructure investment. The company’s shares rose 4.7% % after the announcement, and Metos executives say Muse Wonder will help the company’s personal AI agent, Muse, gain greater functionality. It shows that Meta is aiming to differentiate itself not just in the coding agent market, but also in a wider range of personal and professional tasks.
Conclusions
Meta Muse Wonder and Muse Wonder Full represent the company’s efforts to reach the frontier level of AI, competing with ChatGPT and Claude. While some of the benchmark data is internal and not always under control, the model’s new features – Active Memory and Active Focus – offer the potential to handle complex, multi-step tasks more efficiently. Availability via API and different pricing plans allow both developers and businesses to try this model out, but the final results will depend on independent tests and real-world usage experience.
—
Meta Muse Wonder AI – this is a new opportunity for both developers and companies seeking to automate complex processes, but it is important to monitor the results of independent research and security reports to assess the true value of the technology.
Sources
- Naturalnews.com - Meta Releases Muse Wonder AI Model, Claims It Can Equal ChatGPT and Claude
- SiliconANGLE News – Meta releases Muse Wonder AI models built to run personal and professional agents
- Forkast.news – Meta Ships Muse Wonder, Its First Frontier Model — and Leaves the Research Blog Empty
- Business Standard - Meta says new Muse model reaches AI frontier after aggressive push
- Asianetnews.com - Meta unveils Muse Wonder AI models for complex professional tasks
