Meta releases Muse Wonder and Muse Wonder Full, with Full limited to Muse Code subscribers

AIAlexandr Wang says Meta's Muse Wonder model is state of the art for personal agents, judged by an internal Personal Superintelligence Index (PSI) rather than external benchmarks. He credits its strengths in Active Memory and Active Focus, and says it has strong general capabilities, which he calls "big model smell."

AIDesign Arena says Meta's Muse Wonder model is now available on its platform. Meta quotes its own launch: Muse Wonder and Muse Wonder Full are frontier models for personal and professional agentic tasks. Muse Wonder is available through the Meta Model API and Muse Code, Muse Wonder Full to Muse Code subscribers, and Muse Wonder will soon power Meta's Muse personal agent.

AIMeta's Alexandr Wang announces Muse Wonder, a model built to power personal agents. It claims the best memory across days, weeks, and months, state-of-the-art results in finance and legal, and strong performance in health.

AIArtificial Analysis has published full Intelligence Index results for Muse Wonder Full (max) and Muse Wonder (max). The post itself contains no scores or further details.

AIMuse Wonder Full (max) scores 30 on AA-Omniscience, up 5 points, after its hallucination rate fell from 33% to 23%. That rate is the second-lowest among models scoring 45 or more on the Intelligence Index, behind only Gemini 4 Argon (high) at 15%. Its knowledge score still trails Claude Opus 5.5 (46), GPT-6 Astra (43), and Gemini 4 Argon (42).

AIMuse Wonder Full (max) places first on CritPt and GDP.pdf at 33% each, narrowly ahead of rivals scoring 31-32%. It takes second on GDPval-AA v2.1 with 1857 Elo, within the confidence intervals of Claude Opus 5.5 and Claude Sonnet 5.5. Its largest gain over Muse Spark 1.3 is on Terminal-Bench 4.0, rising from 33% to 57%.

AIArtificial Analysis says Meta released Muse Wonder and Muse Wonder Full, scoring 52 and 56 on its Intelligence Index v4.3 at max reasoning effort. Both beat Muse Spark 1.3 at 48, with the largest gains in agentic evaluations such as Terminal-Bench 4.0. Muse Wonder costs $4.52 per Index task, which is higher than most similarly intelligent models, and Muse Wonder Full has no per-token price yet.

AIMeta has released Muse Wonder, a model family whose top version, Full, beats Claude Opus 5.5 on 8 of 14 published benchmarks. Full scores 57.5 against 46.3 on SWE-Marathon, an ultra-long engineering test, and 79.0 against 74.2 on DeepSWE 1.1, where Meta says the runs were independent and third-party. On Harvey's legal agent benchmark, Full scores 13.1 against Opus 5.5's 4.2, crediting only fully finished tasks without hallucination.

AIMeta says Muse Wonder is its most capable model yet, built for complex personal and professional agentic tasks. The company says it delivers frontier coding performance, advanced computer use, and handles long-horizon work for developers' most demanding tasks.

AIMeta's Muse Wonder Full scores 2 points behind Claude Opus 5.5 on the Artificial Analysis Intelligence Index. It reaches 79.0% on DeepSWE v1.1, ahead of Astra (74.1%) and Opus (74.2%), and beats both on SWE-Marathon. The post author asks Meta to open-source Muse Wonder, or at least Muse Spark 1.3.

AIMeta launched the Muse Wonder series, with Muse Wonder and Muse Wonder Full, which it describes as its most capable models yet. According to TestingCatalog, Muse Wonder Full scores second on the Artificial Analysis Index, 2 points behind Claude Opus 5.5, and scores 79 on DeepSWE. Muse Wonder is available through the Meta Model API and Muse Code, while Muse Wonder Full is limited to Muse Code subscribers.

AIAlexandr Wang, who owns the account and is tied to Meta and Llama, says Muse Wonder is the best model in the world for personal and professional agents. He says Meta evaluated the model against top models on agentic and coding tasks, but the post gives no benchmark scores.

AIMeta says it trained the Muse Wonder model family to lead on alignment and safety, with robustness to adversarial instructions and careful handling of private information. The post frames these two capabilities as necessary for delivering personal superintelligence.

AIMeta says Muse Wonder is built for long-running, multi-step agent work through two features. Active Memory carries context across conversations and surfaces it when relevant. Active Focus keeps the models pursuing a goal through blockers by trying alternative approaches until the task is complete.

AIMeta says its Muse Wonder series achieves state-of-the-art results across key benchmarks for personal and professional agents, memory, and coding. The models, Muse Wonder and Muse Wonder Full, are built to apply specialized knowledge to everyday work and life problems, such as understanding a lease or interpreting financial or health information. Meta says its evaluations include tasks in law, finance, and health that test this ability.

AIMeta announces the Muse Wonder series, with Muse Wonder and Muse Wonder Full, which it describes as its most capable models yet. Muse Wonder is available now through the Meta Model API and Muse Code, while Muse Wonder Full is available today only to Muse Code subscribers. Muse Wonder will also soon power experiences in Meta's personal agent product Muse, rolling out first to subscribers.

AIAlexandr Wang announces Muse Wonder, the result of over a year of research and infrastructure work in Meta Superintelligence Labs. He says Meta has scaled its models across pre-training and RL since Muse Spark shipped in April, reaching the frontier on the Artificial Analysis Intelligence Index.

AIAndrew Curran says Muse Wonder, also called Watermelon, has arrived. The post gives no further details on the model's capabilities, release, or availability. Background from ChrisGPT names it Muse Wonder 1.0.

AIMeta announces Muse Wonder, its new frontier model, which it says is state-of-the-art at personal and professional agents. Meta describes Muse Wonder as its largest and most capable model yet, rivaling the biggest frontier models.

AIMark Zuckerberg says Meta is releasing two models today, Muse Wonder and Muse Wonder Full, following Muse Spark as the first family of Meta Superintelligence Labs. He describes Muse Wonder as a much larger and more advanced class of models that brings Meta to the frontier, with Muse Wonder for fast frontier performance and Muse Wonder Full for the full version.

AIMeta announces Muse Wonder, a new family of frontier models with state-of-the-art performance for personal and professional agentic work. The post gives no benchmark scores, parameter counts, prices, or availability dates.

AITesting Catalog reports that Meta is rolling out Muse Wonder and Muse Wonder Full on the Meta Model API platform. Muse Wonder is described as state-of-the-art at real-world coding and long-running tasks, with tool calls in parallel and sub-agent handoffs. Muse Wonder Full, the most capable model at max reasoning, is available only to Muse Code subscribers and not through the API.