Meta returns to the race for the most powerful AI models with Muse Wonder and Muse Wonder Full. Launched on October 12, both models display a context window of one million tokens. On several benchmarks published by Meta, Wonder Full is ahead of GPT-6 Astra and Claude Opus 5.5. Independent tests are more nuanced: Anthropic remains ahead in the general ranking.
In brief
- Meta launches Muse Wonder and Wonder Full with one million context tokens.
- Wonder Full outperforms GPT-6 Astra on several tests presented by Meta.
- Artificial Analysis gives it 56 points, which places it among the best current AI models.
Muse Wonder puts Meta back in the AI race
Meta presents Muse Wonder as an important step towards “personal superintelligence” defended for several months by Mark Zuckerberg. The launch comes after a much less comfortable period for the group. In July, Zuckerberg still recognized that Meta was moving too slowly in AI, despite considerable investments and a profound reorganization of its teams.
Muse Wonder changes the picture a little. The model has a context window of one million tokens, enough to manipulate very large quantities of text during a single task. Meta also introduces two central functions: Active Memory and Active Focus.
Active Memory allows the AI to retain some useful information between different conversations. Active Focus should allow him to stay longer on a complex problem, to try several approaches and to continue his work when a first method fails.
The objective therefore goes beyond the simple generation of responses. Meta wants agents who can work for multiple steps on code, documents, or business tasks. Muse Wonder is available in the Meta Model API and Muse Code. The Wonder Full version is currently reserved for Muse Code subscribers.
Meta’s AI beats Astra on several benchmarks
Meta accompanies this release with several particularly high results. On its Personal Superintelligence Index, Wonder Full scores 59.1 points. Claude Opus 5.5 reached 55.2 points and GPT-6 Astra 54.5.
The gap increases on certain specific tasks. SWE-Marathon measures the ability of an AI to work for a long time on programming problems. Wonder Full obtains 57.5, compared to 46.3 for Claude Opus 5.5 and 42.5 for GPT-6 Astra.
Meta’s model also dominates the Legal Agent Benchmark used to test professional legal tasks. Its score reaches 13.1, compared to 8.6 for Astra and 4.2 for Opus 5.5. Memory is another strong point. On MemoryBench, Wonder and Wonder Full are around 85.5. Astra reaches 81.4.
Meta now has a much more tailored model for Muse, its AI agent capable of browsing the web, filling out forms and making purchases. One clarification remains important: these first figures come from evaluations published by Meta. They don’t mean that Wonder Full is automatically better than Astra in every area.
Independent tests put Claude back ahead of Meta
Artificial Analysis provides another point of comparison. His Intelligence Index gives 56 points to Muse Wonder Full and 52 to Muse Wonder. Progress remains significant. Muse Spark 1.3 still only obtained 48 points six weeks ago.
Wonder Full thus enters the group of the most efficient AI models of the moment. Anthropic nevertheless maintains the advantage on this independent assessment. Claude Opus 5.5 achieves a higher score than the Meta model.
Wonder also shows some weaknesses. Artificial Analysis notes in particular a delay in general knowledge tests. Wonder Full obtains 30 on AA-Omniscience, against 43 for GPT-6 Astra and 46 for Claude Opus 5.5.
The model stands out more on agents, scientific reasoning and certain programming tasks. The price provides another element of comparison. Meta charges Muse Wonder $4 per million tokens for input and $20 for output. GPT-6 Astra costs $10 input and $50 output.
Meta therefore does not win all benchmarks. But the gap has narrowed considerably. A few weeks after the release of GPT-6 Astra, presented by OpenAI as a major new step in its AI, Zuckerberg now has a model capable of competing directly with it in several categories. After months of running behind OpenAI and Anthropic, Meta is finally returning to the top group.
Maximize your Tremplin.io experience with our ‘Read to Earn’ program! For every article you read, earn points and access exclusive rewards. Sign up now and start earning benefits.
