Meta is no longer giving its best AI away — Muse Wonder tests whether a closed, paid model can sit beside Anthropic and OpenAI at the frontier
October 13, 2026 — Meta launched Muse Wonder and Muse Wonder Full on October 12, 2026, calling them, according to the company, its most capable models yet. The pair marks Meta's first frontier-class release, built for personal and professional agent work, coding, memory and long multi-step tasks.
Image: A smartphone and laptop showing a glowing AI assistant, with luminous threads linking conversations. Credit: AI-generated illustration / Red Eye Eagle News, original illustration.
Editor's Highlight: The launch is a strategy shift as much as a model release. After years centred on free, open AI, Meta is now selling its strongest system as a closed, paid service — priced exactly like a leading rival — while reserving the best version for subscribers only.
Table of Contents
- What Meta Released on October 12
- How the Two Models Differ
- What Independent Scores Show
- Active Memory and Active Focus
- Pricing and the End of Free Frontier Access
- Limits Meta Itself Lists
- Bottom Line
- FAQ
- Sources
Key Facts
| Detail | Information |
|---|---|
| Launch date | October 12, 2026 |
| Models | Muse Wonder and Muse Wonder Full |
| Builder | Meta Superintelligence Labs, according to the company |
| Wonder access | Meta Model API and Muse Code |
| Wonder Full access | Muse Code subscribers only; not on the pay-as-you-go API |
| API price | $4 per million input tokens; $20 per million output tokens |
| Independent score | Artificial Analysis: Wonder Full 56, Wonder 52 |
| Open weights | None announced; both are hosted, closed services |
What Meta Released on October 12
Meta describes the Muse Wonder family as frontier models for personal and professional agentic tasks.
The standard Muse Wonder model is available now through the Meta Model API and Muse Code. Meta says it will also power the company's Muse personal agent, with subscribers getting access first.
Newsroom Focus: This replaces the earlier Muse Spark as Meta's headline model. Spark was positioned as a lighter, faster system; Wonder is positioned on capability first.
How the Two Models Differ
Wonder Full is the larger, higher-scoring model, and it is limited to Muse Code subscribers.
According to published plan details, Wonder Full is included in Muse Code's Power Usage plan at $50 a month and Maximum Usage plan at $200 a month. It is not available through the pay-as-you-go API.
| Model | Where to get it | Intelligence Index |
|---|---|---|
| Muse Wonder | Meta Model API, Muse Code | 52 |
| Muse Wonder Full | Muse Code subscribers only | 56 |
| Claude Opus 5.5 | Anthropic | 58 |
Both Meta models can process up to one million tokens of context, according to reports of the launch specifications. They accept text, images, video and audio as input, with text as output.
What Independent Scores Show
Meta says, in its own testing, that Wonder Full performs above Anthropic's Claude Fable 5.1 on selected tests and close to Claude Opus 5.5. Those comparisons are company claims and should be read as such.
The independent picture is close but not identical. Artificial Analysis scored Wonder Full at 56 and Wonder at 52 on its Intelligence Index.
That places Wonder Full behind Claude Opus 5.5 at 58, and ahead of Claude Fable 5.1 and OpenAI's GPT-6 Astra, each at 53.
Analysis Note: One index is a useful starting point, not a final verdict. Teams should test any model on their own real tasks — the same inputs, the same checks — before switching workloads or budgets.
Active Memory and Active Focus
Meta is pitching two features as the core of the agent offer. The first is Active Memory, which the company says carries important context across conversations and retrieves details relevant to a new task.
The second is Active Focus. According to Meta, it helps the model keep working toward a difficult goal, try alternative approaches when blocked, and ask for clarification or confirmation before important actions.
Pricing and the End of Free Frontier Access
For developers, Muse Wonder costs $4 per million input tokens and $20 per million output tokens through the Meta Model API. Cached input is listed at $0.20 per million tokens, and web search at $2.50 per 1,000 queries.
That price matches leading rivals rather than undercutting them, and is a step up from Muse Spark pricing reported by Artificial Analysis.
Neither model was announced as an open-weight release — a notable change for the company behind downloadable Llama weights.
Practical Takeaway: Budget on cost per finished task, not cost per token alone. A model that uses fewer tokens to complete the same job can be cheaper in practice, even at the same list price.
Limits Meta Itself Lists
Meta's model card, as reported at launch, lists known limitations. They include inaccurate responses, errors in multi-step reasoning, and uneven performance across languages.
Muse Code subscriptions default to a contributor setting that allows prompts, code, files and responses to be used to train future models, according to Meta's documentation; subscribers can turn it off.
Pay-as-you-go API access uses a standard mode in which, Meta says, submitted content is not used to train its models.
Bottom Line
Meta has entered the frontier race with a credible, independently scored model family — and it has done so on commercial terms that look much more like its rivals than like its open-source past.
Watch next for a dated Muse consumer rollout, any open-weights decision, and independent tests on coding and long agent work. Until then, Muse Wonder is competitive at the frontier, not proven above it.
FAQ
What is Muse Wonder?
It is a family of AI models Meta released on October 12, 2026. It includes Muse Wonder, available through the Meta Model API and Muse Code, and the larger Muse Wonder Full, limited to Muse Code subscribers.
How much does Muse Wonder cost?
Through the Meta Model API, Meta lists $4 per million input tokens and $20 per million output tokens. Wonder Full is included in Muse Code plans reported at $50 and $200 a month, and is not sold through the pay-as-you-go API.
Is Muse Wonder open source?
No open-weight or open-source release was announced. Both models are hosted services, which is a change from Meta's earlier Llama approach.
How does it compare with Claude and GPT models?
Artificial Analysis scored Wonder Full at 56 and Wonder at 52 on its Intelligence Index. That is behind Claude Opus 5.5 at 58, and ahead of Claude Fable 5.1 and GPT-6 Astra at 53 each. Meta's own tests give a more favourable picture, so treat company charts as claims.
When will Muse Wonder reach the Muse app?
Meta says Muse Wonder will power its Muse personal agent, starting with subscribers. The company gave no rollout date at launch.
Sources
- Seoul Economic Daily (Reuters-Yonhap) — Meta launches paid AI model Muse Wonder to speed monetization: https://en.sedaily.com/international/2026/10/13/meta-launches-paid-ai-model-muse-wonder-to-speed
- Okay News — Meta launches Muse Wonder preview: prices and access: https://www.okaynews.com/meta-muse-wonder-preview-api-pricing-october-2026/
- AIStockWire — Meta says Muse Wonder models reach the AI frontier: https://aistockwire.com/blog/meta-muse-wonder-full-frontier-ai-models-openai-anthropic-october-2026
- Oossa — Meta launches Muse Wonder models, hitting top AI benchmark scores: https://oossa.com/en/meta-launches-muse-wonder-models-hitting-top-ai-benchmark-scores
- Image: AI-generated illustration created for Red Eye Eagle News (original illustration, no external photo source).
Join the conversation
Comments
Post a Comment