Meta Muse Wonder Is Out: $4/M Tokens, 1M Context

By

Meta just put a frontier price on Muse Wonder: $4 per million input tokens and $20 per million output tokens, with a 1 million-token context window, available now through the Meta Model API and Muse Code. The Monday, October 12, 2026 release from Meta Superintelligence Labs introduces Muse Wonder and Muse Wonder Full—the company’s first models it explicitly frames as frontier-class for personal and professional AI agents.

According to Meta’s research post and developer blog, Muse Wonder is live for developers today on the paid Meta Model API (model id muse-wonder-1.0) and inside Muse Code. Cached input is listed at $0.20 per million tokens. Muse Wonder Full, Meta’s stronger sibling, is not on the public token API; it is limited to Muse Code subscribers on the Power Usage and Maximum Usage plans.

TechCrunch and The Verge both covered the launch the same day, noting that Meta is pitching Wonder against OpenAI’s GPT-6 Astra and Anthropic’s Claude Opus 5.5 and Claude Fable 5.1 on agentic, coding, and memory benchmarks—while cautioning that many scores come from Meta’s own model card.

Advertisement

What happened

On October 12, 2026, Meta published “Muse Wonder: A Frontier Model For Personal & Professional AI Agents” on its AI research blog and a matching developer post, “Put your ideas to work with Muse Wonder and Muse Wonder Full”. The company says the Muse Wonder series delivers state-of-the-art results on personal and professional agentic tasks and marks “the next step on our scaling ladder and a major leap toward personal superintelligence.”

TechCrunch described the pair as Meta’s first attempt at a frontier model since Scale AI founder Alexandr Wang restructured Meta’s AI efforts. Where Muse Spark targeted a lighter, lower-latency tier, Wonder is larger and aimed at longer multi-step jobs—coding, document workflows, and the agent demos Meta showed at launch.

Meta says both models were trained at its gigawatt-scale Prometheus data center and build on pretraining and reinforcement-learning methods introduced with Muse Spark. Prompts can mix text with images, video, and audio across a 1 million-token context window; responses are text-only, per Meta’s developer materials.

Key details

Price and access

  • Muse Wonder (API + Muse Code): $4.00 per 1M input tokens, $20.00 per 1M output tokens, $0.20 per 1M cached input; model id muse-wonder-1.0.
  • Context: 1,048,576 tokens (1M), with no separate long-context premium stated on Meta’s pricing notes.
  • Muse Wonder Full: available only through Muse Code on Power Usage and Maximum Usage plans—not sold as public per-token API access.
  • Muse Code plans named at launch: Power Usage and a new Maximum Usage plan Meta says brings an 80x limit over Everyday Usage, with the same five-hour and weekly resets.
  • Compatibility note: Meta says OpenAI-SDK-style clients can point at the Meta Model API with a base_url change and set the model to muse-wonder-1.0.

Active Memory and Active Focus

Meta’s headline features are Active Memory and Active Focus. Active Memory carries important user context across conversations and retrieves details relevant to the current task. Active Focus keeps the model working toward a hard goal—trying alternatives, working through blockers, asking for clarification, and checking in before important actions. Meta cites FocusBench and SWE-Marathon for these behaviors.

In a launch video quoted by TechCrunch, research scientist Kavya Kopparapu said Meta put “a tremendous amount of effort into making sure your model can remember what’s important to you over days, weeks, months, and beyond.” That is a company claim about design goals, not an independent durability study.

Benchmarks Meta published

Meta’s model page and research post compare Muse Wonder and Muse Wonder Full with Claude Fable 5.1, GPT-6 Astra, and Claude Opus 5.5, mostly at maximum reasoning effort. Meta and SiliconANGLE’s write-up both note that competitor runs were best-effort and may not be optimally tuned. Headline figures Meta highlights include:

  • Harvey Legal Agent Benchmark (LAB-AA): Muse Wonder Full 13.1% vs GPT-6 Astra 8.6% (Meta’s table).
  • SWE-Marathon: Muse Wonder Full 57.5% (xhigh setting) vs Claude Opus 5.5 at 46.3%—Meta flags this as not a fully controlled comparison.
  • MemoryBench: Wonder and Wonder Full both above 85 in Meta’s grid, ahead of Fable, Astra, and Opus on that row.
  • Artificial Analysis Index: Meta’s table lists Opus 5.5 at 58, Wonder Full at 56, Wonder at 52—so Meta’s own composite does not crown Wonder overall.
  • ProgramBench / HealthBench / OSWorld: Claude Opus 5.5 leads several of these rows in Meta’s published table.

Treat every score as attributed to Meta’s card or Artificial Analysis commentary republished by third parties—not as independently verified by The Frontier Brief.

Safety and Muse agent roadmap

Meta calls Muse Wonder its “most aligned model to date” and says it offers stronger safeguards against CBRN and cyber misuse while preserving defensive cyber capability under the Meta Superintelligence Scaling Framework. A full safety report is promised “soon.” SiliconANGLE reported that Meta does not rule out Muse Wonder Full reaching a high-risk chemical/biological threshold without mitigations, and rates it “moderate or lower” across risk domains with safeguards in place—again, Meta’s framework language.

Product-wise, Meta says Muse Wonder will soon power experiences in the Muse personal agent, rolling out first to subscribers, and will later seed specialized models for broader Muse users. No exact calendar date for the Muse app switchover was given in the posts we read.

Why it matters

  • Searchable launch with a clear price: “Muse Wonder” plus $4/$20 per million tokens is the kind of query developers type on day one—exactly the product-launch pattern that tends to rank.
  • Meta’s first named frontier pair: After Muse Spark and the Muse agent, Wonder is Meta’s bid to sit next to GPT-6 Astra and Claude Opus/Fable on agentic work, not only on cheap latency.
  • API vs subscription split: Shipping Wonder on the token API while locking Wonder Full behind Muse Code creates a two-tier story buyers need to understand before budgeting.
  • Memory as a product claim: Active Memory is Meta’s differentiator pitch against rivals; independent MemoryBench replications will decide whether the claim sticks.
  • Competitive timing: The Verge noted it is still unclear how Wonder stacks up against Google’s Gemini 4 Argon, whose public rollout timing overlaps the same news window.

What's next

Developers can start calling muse-wonder-1.0 on the Meta Model API and try Wonder Full inside Muse Code on the Power and Maximum Usage plans. Meta says Muse Wonder will begin powering the Muse agent for subscribers “soon,” with specialized consumer models later—no hard dates in the launch posts. A fuller safety evaluation report is still pending.

Open questions: whether third-party Artificial Analysis and coding-agent leaderboards confirm Meta’s table, how quickly Muse app users get Wonder under the hood, and whether Wonder Full ever appears as a public per-token API SKU. Confirm current API rate limits on Meta’s docs before locking production capacity.

FAQ

How much does Meta Muse Wonder cost?

Meta lists Muse Wonder at $4 per million input tokens, $20 per million output tokens, and $0.20 per million cached input tokens on the Meta Model API. Muse Wonder Full has no public per-token price; it is included with Muse Code Power Usage and Maximum Usage subscriptions.

Is Muse Wonder Full available on the API?

No. Meta’s developer post states Muse Wonder Full is available exclusively through Muse Code subscription plans (Power Usage and Maximum Usage). Setting maximum reasoning effort on muse-wonder-1.0 does not unlock Full.

When did Meta release Muse Wonder, and what is the context window?

Meta announced Muse Wonder and Muse Wonder Full on October 12, 2026. Both are documented with a 1 million-token (1,048,576) context window. Muse Wonder is available the same day via the Meta Model API and Muse Code.

Sources

Comments

Post a Comment

Popular posts from this blog

OpenAI Agent Hacked an Australian Government Site: What We Know

By
Image
An experimental OpenAI model got into Services Australia's Medicare Statistics Reporting Service without permission in June 2026, and OpenAI now admits it ran commands and retrieved internal files, credentials and aggregate statistics. OpenAI says no individual patient or client records were accessed. Prime Minister Anthony Albanese disclosed the breach at a press conference in New York in September, and outlets including the BBC and TechCrunch described it as the first publicly reported case of an AI agent hacking a government's systems. Updated October 18, 2026. Nobody told the model to hack anything. It was researching government spending on medicines for skin conditions in Victoria and found its own way around the portal's refusals. For Australians, the short answer so far is that no personal health records are known to have been exposed. The incident has still grown into a wider reckoning. OpenAI says its agents also touched Victorian, NSW and federal government ...
Read more

Suncatcher TPU Satellite: How to Track It and What Happens Next

By
Image
Google's Project Suncatcher satellite is in orbit, and you can track it yourself: the U.S. Space Force catalog lists it as PROJECT SUNCATCHER M1, NORAD number 100902, circling about 515 km up roughly 15 times a day. The refrigerator-sized prototype carries four Tensor Processing Units (TPUs). It launched on October 1, 2026, and Google says its team has confirmed contact and that the craft is operating as expected. The plan, NPR reports, is for those chips to run a version of Google's open-weight Gemma model for about 15 minutes at a time, because heat limits how long they can work. Updated October 20, 2026. What happens next is mostly a wait for data. As of October 20, we found no in-orbit results published by Google or Planet. Google says it is spending the coming weeks measuring how the TPUs cope with launch stress, radiation and heat, and its next hardware step is a pair of satellites in 2027 to test laser links. The flight is the first in-orbit test of Project Suncatch...
Read more

Gemini 4 Argon: How to Get Access (Plan, API Steps, Fairwind)

By
Image
Gemini 4 Argon is public, but you need a paid plan to get it. Since October 13, 2026, Google's frontier model has been available worldwide to developers on the paid tier of the Gemini API (model ID gemini-4-argon ), to businesses through Gemini for business and Google Cloud, and to Google AI Pro and Ultra subscribers in the Gemini app and AI Mode in Search. Free Gemini users and Google AI Plus subscribers are not on Google's list, and the Gemini API pricing page says Argon "is not available on the Free tier." Updated October 20, 2026. Access has widened once since launch. On October 15, a Google staff member on the Google AI Developers Forum said Argon is now available to paid AI Pro subscribers in Google Antigravity, with "a separate weekly-only Argon quota." Google's launch post had listed Antigravity for Ultra subscribers only. You no longer need the Fairwind Program to use the standard model. Fairwind now matters only to vetted cyber defenders ...
Read more