Meta Superintelligence Labs prices Muse Wonder at $4 (€3.56) per million input tokens and keeps the larger Muse Wonder Full for paid Muse Code plans.
Meta launched Muse Wonder, the company’s first frontier AI model, on 12 October 2026 alongside the larger Muse Wonder Full. Meta Superintelligence Labs priced the smaller model at $4 (€3.56) per million input tokens and kept the larger model for paid plans. TF breaks down the scores, prices, and trade-offs.
What’s Happening & Why It Matters
The Launch
Meta Superintelligence Labs announced both models on 12 October. Alexandr Wang, who leads the lab, calls the family the company’s largest and most capable. Both models are proprietary and have no downloadable weights.
The models take text, images, audio and video, and return text. The context window holds about 1 million tokens. Meta trained them on its Prometheus data centre in Ohio, using large-scale pre-training and reinforcement learning. Euronews says the smaller model suits daily tasks, and the larger model suits harder work.
Meta adds two agent features. Active Memory keeps personal and professional context across chats. Active Focus keeps an agent working after failed attempts and asks permission before consequential actions.
The Scores
Artificial Analysis rates Muse Wonder Full at 56 and Muse Wonder at 52 on its Intelligence Index. Anthropic’s Claude Opus 5.5 scores 58, and Claude Sonnet 5.5 scores 56. OpenAI’s GPT-6 Astra and Google’s Gemini 4 Argon each score 53.
Meta’s table gives Wonder Full a lead in two coding tests. Wonder Full scores 79.0 on DeepSWE 1.1 against 74.2 for Opus 5.5. Wonder Full scores 57.5 on SWE-Marathon against 46.3. Opus 5.5 leads on Terminal-Bench 4.0 (59.6 to 57.1) and HealthBench Pro (65.6 to 62.3).
Treat the table with care. VentureBeat says Meta mixes internal tests, vendor scores and third-party results. Some coding comparisons used different tools. No source reports speed or latency figures.
Pricing and Access
The Muse Wonder API costs $4 (€3.56) for input and $20 (€17.80) for output per million tokens. Claude Opus 5.5 costs the same. GPT-6 Astra costs $10 (€8.90) for input and $50 (€44.50) for output per million tokens. Artificial Analysis puts an average Index task at $4.52 for Wonder, $5.98 for Opus 5.5 and $3.26 for GPT-6 Astra.
Muse Wonder Full has no API price. Developers get the model through two Muse Code plans. Power Usage costs $50 (€44.50) a month and gives 20 times the Everyday allowance. Maximum Usage costs $200 (€178) a month and gives 80 times the Everyday allowance.
Safety and Data
Meta’s model card covers cybersecurity, biological and chemical risks, and autonomous actions. Meta recommends human confirmation for consequential agent actions. Mark Zuckerberg says Meta delayed the release by several months for safety work.
Microsoft’s Satya Nadella and Brad Smith urge the same caution, as reported in Microsoft’s Nadella and Smith Call for an AI Emergency Brake. Muse Code subscriptions default to Contributor mode, where Meta may use prompts, code and files to train future models. Users can switch the mode off for future submissions. The Wonder API uses standard mode, with no training on submitted content.
My take: price is the story. Meta matched Anthropic’s Opus rate for a model scoring 6 points lower. Meta kept the stronger model behind a subscription.
TF Summary: What’s Next
Developers will test Wonder on real agent work over the coming weeks. Meta has to say whether Wonder Full reaches the API. Artificial Analysis and others will publish more runs. Rivals will respond on price.
MY FORECAST: Meta will open Muse Wonder Full to the API within six months. Independent tests will place Wonder Full below Meta’s table on at least one coding benchmark. Anthropic and OpenAI will cut prices on their top models by mid-2027. Regulators will ask Meta for incident-reporting terms.
Related Stories
- Microsoft’s Nadella and Smith Call for an AI Emergency Brake
- TF Cybercrime Round-Up: 6-11 October 2026
- RAMageddon Hits MacBook Air Supply as Chip Shortage Deepens
- Samsung Posts Record $80 Billion AI Profit
- Denmark Bill Would Ban Deepfakes of Faces and Voices