Meta releases Muse Wonder and Muse Wonder Full agent models
Meta released two new AI models, Muse Wonder and Muse Wonder Full, which it says achieve state-of-the-art performance on personal and professional agent tasks. Muse Wonder is available through the Meta Model API and Muse Code, while Muse Wonder Full is limited to Muse Code subscribers. The models target expert domains including law, finance, and health, and feature Active Memory and Active Focus. Meta's performance claims have not been independently verified in the provided sources.
IllustrationEditorial responsibility
- No named human review is recorded for this page.
- Reports are grouped by semantic similarity and deterministic rules. Language models may assist titles, summaries, translation and cross-source analysis; the page reads the event directly, while its address stays stable when the title changes.
- Summary covers the current reports
Cross-source coverage
Reporting timeline
Meta's Muse Wonder Tops Opus 5.5 and GPT-6 Astra on Benchmarks
A post on X claims that Meta's Muse Wonder outperforms both Opus 5.5 and GPT-6 Astra on every listed benchmark. The widest margin is reported on SWE-Marathon, a test of whether an AI agent can independently build an entire software project, such as a C compiler or a Slack clone, over 2 to 10 hours, with a task counted only if every hidden check passes. According to the post, Muse Wonder full scores 57.5 on that benchmark versus 46.3 for Opus 5.5. The post offers no further methodology, dates, or independent verification, and the benchmark results are presented as the poster's claim rather than a confirmed announcement.
Read sourceMeta's Superintelligence Lab releases Muse Wonder and Muse Wonder Full models
Meta's Superintelligence Lab released two new state-of-the-art models on Monday, named Muse Wonder and Muse Wonder Full. According to the source item, this marks the company's first attempt at a frontier model. The announcement names both models and characterizes them as state-of-the-art, but provides no further details on capabilities, benchmarks, availability, or pricing. The post links to an external URL for additional information. No attributed opinions, forecasts, or third-party reactions are included in the source text.
Read sourceMeta Releases Muse Wonder and Muse Wonder Full Models for Personal and Professional Agents
Meta announced the release of its Muse Wonder series of models, which it says achieve state-of-the-art performance on personal and professional agent tasks. According to the announcement, Muse Wonder is now available through the Meta Model API and Muse Code, while Muse Wonder Full is offered to Muse Code subscribers. The company did not disclose benchmark details, pricing, or technical specifications in the provided text. The claim of SOTA performance is attributed to Meta and has not been independently verified in this item.
Show 8 older updatesHide older updates
Meta releases Muse Wonder and Muse Wonder Full, scoring 52 and 56 on Intelligence Index
Meta released two new AI models, Muse Wonder and Muse Wonder Full, which scored 52 and 56 respectively on the Artificial Analysis Intelligence Index v4.3 at max reasoning effort, up from 48 for Muse Spark 1.3 (max) just six weeks earlier. Muse Wonder Full improved across all ten evaluations, with large gains in agentic tasks such as Terminal-Bench 4.0 (+24 points to 57%), GDPval-AA v2.1 (+176 Elo to 1857), and AA-Briefcase (+148 Elo to 1729). Both models use about 54k output tokens per task, roughly 10% less than Muse Spark 1.3, and Muse Wonder Full uses 16% fewer input tokens than Muse Wonder. Muse Wonder costs $4.52 per Intelligence Index task, higher than several peers but lower than some Claude models. Muse Wonder is available on Meta's first-party API and Muse Code, while Muse Wonder Full is limited to Muse Code users.
Read sourceMeta Releases Muse Wonder, Beating Claude Opus 5.5 on 8 of 14 Benchmarks
Meta has released Muse Wonder, an AI model whose top version outperforms Claude Opus 5.5 on 8 of 14 published benchmarks, according to the source post. The post describes the release as a major milestone for Mark Zuckerberg and says Muse Wonder is ahead of GPT-6 Astra and Claude Fable 5.1 on long-running coding, legal work, and business automation. The widest reported gap is on SWE-Marathon, an ultra-long engineering test, where the Full version scores 57.5 versus 46.3 for Opus 5.5. Full also reportedly leads DeepSWE 1.1, a long-horizon coding test, 79.0 to 74.2, in what Meta says were independent third-party runs. On Harvey's legal agent benchmark, which credits only fully completed tasks without hallucination, Full scores 13.1, more than triple Opus 5.5's 4.2.
Read sourceMeta Details Muse Wonder AI Model With Active Memory and Active Focus
Meta provided a closer look at its Muse Wonder AI model, stating that significant effort went into ensuring the model can remember what matters to users over days, weeks, months, and beyond. Meta calls this capability active memory, which it says allows Muse Wonder to recall user habits, preferences, and previous conversations over extended periods. The company also introduced active focus, training the model on real-world failure scenarios so it can anticipate problems and pursue alternative solutions when an agent gets stuck. One demonstrated example showed Muse Wonder continuously researching rental listings across multiple sources, filtering them according to personal constraints, and autonomously contacting property owners. The post's author noted surprise that the video was not shown on X.
Read sourceMeta's Muse Wonder Full Ranks Second on Artificial Analysis Index, 2 Points Behind Claude Opus 5.5
Meta announced that its Muse Wonder Full model placed second on the Artificial Analysis Index, just 2 points behind Claude Opus 5.5, and achieved state-of-the-art results on DeepSWE with a score of 79 points. The post states that Muse Wonder capabilities cover Active Memory, Active Focus, Frontier Coding, and more. Muse Wonder is now available on the Meta Model API and in Muse Code, while Muse Wonder Full is launching exclusively for Muse Code subscribers. A quoted description says Muse Wonder and Muse Wonder Full apply specialized knowledge to problems people encounter at work and in daily life, from understanding a lease to making sense of financial or health information. The post also asserts that Meta is among the top AI labs now.
Read sourceMeta Releases Muse Wonder and Muse Wonder Full Agent Models
Meta announced the release of two new models, Muse Wonder and Muse Wonder Full, which it says achieve state-of-the-art performance on personal and professional agent tasks. According to the announcement, the models target expert domains including law, finance, and health, and offer frontier coding performance. They include Active Memory, which retains information across conversations, and Active Focus, which can switch paths when errors occur and asks for confirmation when needed. On safety, Meta says the models are trained not to share private information without permission and to refuse harmful requests. The official release notes list key capabilities for real-world tasks in law, finance, and health, giving readers a sense of the models' positioning and how to access them.
Read sourceMeta Introduces Muse Wonder and Muse Wonder Full Frontier Models
Meta announced the Muse Wonder series, comprising two frontier models, Muse Wonder and Muse Wonder Full, which the company describes as its most capable models yet, built for state-of-the-art performance on personal and professional agentic tasks. Muse Wonder is available today through the Meta Model API and Muse Code, while Muse Wonder Full is available today for Muse Code subscribers. The company also said Muse Wonder will soon power experiences in its personal agent product, Muse, rolling out first to subscribers. No benchmark figures, pricing, or model specifications were disclosed in the post.
Read sourceMeta says Muse Wonder Full beats OpenAI and Anthropic on agent tests
Meta claims its Muse Wonder Full models outperform leading OpenAI and Anthropic models on several tests of sustained agent work. According to the post, Muse Wonder Full outperforms all other models on DeepSWE, including Opus 5.5. On SWE-Marathon, its reported score is 57.5%, compared with 46.3% for Claude Opus 5.5 and 42.5% for GPT-6 Astra. The post also says Active Memory retrieves relevant context across conversations, while Active Focus helps the models work through blockers. Wonder is available through Meta's API and Muse Code, and Full is available to Muse Code subscribers. The post frames the results as a significant competitive development and expects OpenAI and Anthropic to respond quickly.
Read sourceMeta Rolls Out Muse Wonder and Muse Wonder Full on Meta Model API Platform
Meta is rolling out two new models, Muse Wonder and Muse Wonder Full, on its Meta Model API platform. According to the announcement, Muse Wonder Full is described as Meta's most capable model, running at maximum reasoning for the hardest and longest-running work, and is available to Muse Code subscribers rather than on the API. Muse Wonder is described as state-of-the-art at real-world coding and built for long-running tasks; it plans work, calls tools in parallel, hands parts off to sub-agents, and recovers from errors so complex jobs finish. The post does not specify pricing, release dates, or benchmark results.
Read source