Meta launches Muse Wonder AI models and claims parity at the frontier
Meta has released Muse Wonder and Muse Wonder Full, claiming state-of-the-art agent capabilities as CEO Mark Zuckerberg declared the company has reached the AI frontier and rejected calls for a development slowdown.
Meta has introduced its latest flagship artificial intelligence models, Muse Wonder and Muse Wonder Full, presenting them as state-of-the-art systems for personal and professional agent workflows. Meta Chief AI Officer Alexandr Wang characterized Muse Wonder as the company's new frontier model, while CEO Mark Zuckerberg told the Financial Times that Meta has caught up with rivals at the frontier of AI, pledging further heavy spending to challenge OpenAI and Anthropic while rejecting calls to slow down development.[1][6][8][12]
Trained at Meta's gigawatt-scale Prometheus data center, the models are built for sustained coding, financial research, and legal tasks. To support long-running workflows, Meta equipped the models with Active Memory, which preserves context across sessions, and Active Focus, designed to help agents navigate blockers by testing alternative approaches before checking in with users.[5][9][13]
Meta reported that Muse Wonder Full outperforms competing frontier models on several sustained-engineering evaluations, posting a 57.5% score on SWE-Marathon compared to 46.3% for Claude Opus 5.5 and 42.5% for GPT-6 Astra, alongside a 79.0 score on DeepSWE 1.1. However, benchmark data shared alongside the launch shows Wonder Full trailing Claude Opus 5.5 on ProgramBench and on the overall Artificial Analysis Index, where Opus 5.5 scored 58 to Wonder Full's 56. Abacus.AI CEO Bindu Reddy also disputed Meta's framing, writing that Wonder Full is nowhere near Astra or Fable levels.[2][3][14]
The standard Muse Wonder model is available through the Meta Model API at $4 per million input tokens and $20 per million output tokens, as well as across Muse subscription tiers and Muse Code. Access to the top-tier Muse Wonder Full model is restricted to Muse Code subscribers paying up to $200 per month.[4][7][11][14]
Key facts
- Meta launched the Muse Wonder and Muse Wonder Full AI models, framing them as state-of-the-art tools for personal and professional agents.
- Mark Zuckerberg told the Financial Times that Meta has caught up to rivals at the AI frontier, rejected calls for an AI slowdown, and pledged further large spending.
- The models were trained at Meta's gigawatt-scale Prometheus data center and include Active Memory and Active Focus features for sustained agent tasks.
- Meta reported Muse Wonder Full scored 57.5% on SWE-Marathon and 79.0 on DeepSWE 1.1, leading Claude Opus 5.5 (46.3% and 74.2) and GPT-6 Astra (42.5% and 74.1).
- Benchmark charts shared by Meta show Claude Opus 5.5 leading the Artificial Analysis Index at 58, ahead of Muse Wonder Full at 56 and Muse Wonder at 52.
- Abacus.AI CEO Bindu Reddy disputed Meta's performance claims, stating that Wonder Full is nowhere near Fable or Astra levels.
- Standard Muse Wonder is available via API at $4 per million input tokens and $20 per million output tokens, while Wonder Full is limited to Muse Code subscriptions costing up to $200 per month.
Sources · 10 sources
- TE
Techmeme@TechmemePost on X ·
Meta introduces Muse Wonder and Muse Wonder Full, saying they deliver state-of-the-art performance on personal and professional agentic tasks (Meta AI Research) (Visit Techmeme dot com for the link and full context!)
Open source - CH
Chubby♨️@kimmonismusPost on X ·
Meta did the unthinkable: Meta says Muse Wonder Full beats leading OpenAI and Anthropic models on several tests of sustained agent work. On DeepSWE it outperforms all other models, even Opus 5.5. On SWE-Marathon, its reported score is 57.5%, versus 46.3% for Claude Opus 5.5 and 42.5% for GPT-6 Astra. Active Memory retrieves relevant context across conversations. Active Focus helps the models work through blockers. Wonder is available through Meta’s API and Muse Code. Full is available to Muse Code subscribers. Meta changed the game. Expect OpenAI and Anthropic to answer ASAP. What a day!!
Open source - DE
DecryptArticle ·
Meta's New Muse Wonder Finally Puts Zuckerberg in the Frontier AI Race Meta's new flagship beats GPT-6 Astra and Claude on memory, legal and marathon coding tests, but independent rankings still put Anthropic's Opus 5.5 on top.
Open source - *B
*Walter Bloomberg@DeItaonePost on X ·
META UNVEILS NEW AI MODEL TO CHALLENGE OPENAI AND ANTHROPIC Meta ($META ) has launched Muse Wonder, a new AI model it claims competes with the world's most advanced systems. Key developments: 🔸Performance: Meta claims Muse Wonder outperforms rival OpenAI and Anthropic models on several benchmarks. 🔸Capabilities: Improved memory, reasoning and expertise in finance, health and law. 🔸Pricing: Coming to Muse's $20 and $100 monthly subscription tiers. 🔸Adoption: Muse has reached nearly 9 million downloads, but growth is slowing and user retention trails ChatGPT. Meta says the model will power its push into personal AI agents, intensifying competition with OpenAI, Google and Anthropic.
Open source - TE
Techmeme@TechmemePost on X ·
Meta says Wonder includes Active Memory and Active Focus, two features designed to help AI systems perform more useful work over extended periods (@inafried / Axios) (Visit Techmeme dot com for the link and full context!)
Open source - EV
Evan@StockMKTNewzPost on X ·
Meta Platforms $META Chief AI Officer Alexandr Wang just posted this: “introducing muse wonder our new frontier model. state-of-the-art at personal agents and professional agents.” https://t.co/8HrCf4UglE
Open source - TE
Techmeme@TechmemePost on X ·
Meta says Muse Wonder is available to Muse app subscribers, via Meta Model API, and Muse Code; Wonder Full is available to Muse Code subscribers (@jonathanvanian / CNBC) (Visit Techmeme dot com for the link and full context!)
Open source - FT
Financial Times@FTPost on X ·
Mark Zuckerberg says Meta has caught up with rivals at the 'frontier' of AI, committing to further spending to challenge Anthropic and OpenAI and rejecting calls to slow down development of the powerful technology. https://t.co/rUYzE9q1aO https://t.co/m2L6BUqUTj
Open source - WS
Wall St Engine@wallstenginePost on X ·
$META LAUNCHES MUSE WONDER FOR PERSONAL AI AGENTS AND CODING Meta introduced Muse Wonder and Muse Wonder Full, calling them its most capable models yet. Both were trained at its gigawatt-scale Prometheus data center. The series combines Active Memory, which carries relevant context across conversations, with Active Focus, designed to keep agents working through complex tasks, trying alternative approaches and checking with users before important actions. The models target personal assistance and professional workflows, including software development and tasks involving legal, financial and health information. Meta reports state-of-the-art results across benchmarks for agents, memory and coding. For developers, the coding capabilities focus on sustained work across projects: tracking requirements, changing multiple parts of a codebase, testing results and making revisions. Muse Wonder is available today through the Meta Model API and Muse Code. Muse Wonder Full is available to Muse Code subscribers. Muse Wonder will also roll out to the Muse app, starting with subscribers, and power artifact generation and other tasks requiring more advanced capabilities. Meta says the models include stronger safeguards for private information, malicious instructions and misuse. A fuller safety report is expected later. https://t.co/4WGS5WXouJ
Open source - FT
Financial Times@FTPost on X ·
Good morning, Asia. While you were sleeping, our most-read article was an exclusive interview with Mark Zuckerberg who pledged further vast AI spending to challenge OpenAI and Anthropic while playing down warnings about the technology’s risks https://t.co/17DNSbDbPV https://t.co/56l2x45b27
Open source - CB
Coin Bureau@coinbureauPost on X ·
⚡️JUST IN: Mark Zuckerberg says Meta has "caught up" with OpenAI and Anthropic at the AI frontier with its new Muse Wonder model, per FT. Meta claims its model beats Claude Opus 5.5 and GPT-6 Astra on select benchmarks, with stronger coding, reasoning and memory capabilities. The $1.8 TRILLION tech company is: - Ramping up AI infrastructure spending - Launching paid AI models for businesses, with premium access at $200/month - Integrating Muse Wonder into its popular AI assistant - Shifting away from its open-source-first strategy Zuckerberg also REJECTED calls to slow AI development, arguing that existing safety guardrails can manage the risks.
Open source - TE
Techmeme@TechmemePost on X ·
An interview with Mark Zuckerberg, who says Meta has caught up with rivals at the frontier of AI with its Wonder model and rejects calls for a slowdown (@mshannahmurphy / Financial Times) (Visit Techmeme dot com for the link and full context!)
Open source - MI
Mint@livemintPost on X ·
Meta has launched Muse Wonder and Muse Wonder Full, its new premium AI models designed for advanced coding, reasoning and complex professional tasks. The models introduce Active Memory and Active Focus, helping the AI retain information across conversations and recover from technical errors. https://t.co/N4OkU7Lkbq
Open source - LD
Lark Davis@LarkDavisPost on X ·
Meta claims new Muse Wonder AI models rival OpenAI, Anthropic. The top-end model, Wonder Full, hit 56 on the Artificial Analysis index, putting it ahead of Astra and Fable (53), though still behind Claude Opus 5.5 (58). Not everyone agrees with Meta's framing, though. Abacus. AI CEO Bindu Reddy pushed back on X, saying the model is nowhere near Fable or Astra level. The catch is access. Wonder Full is locked behind Muse Code subscriptions up to $200 a month. The version developers can actually call via API is standard Muse Wonder, which scored 52, priced at $4 input and $20 output per million tokens.
Open source