Meta now sells a frontier model. Check the training default before your developers sign up.

In brief

Meta Superintelligence Labs released Muse Wonder and Muse Wonder Full on 12 October 2026, describing them as its most capable models to date for "personal and professional agentic tasks". TechCrunch called it Meta's first attempt at a frontier model.

For technology leaders, this adds a serious supplier to the shortlist for coding and agent work. The commercial detail matters as much as the benchmarks: the two variants are sold through different channels on different data terms, and the default on the more capable one is that your content helps train Meta's next models.

What Meta is actually selling

• Muse Wonder on the Model API. Model ID muse-wonder-1.0, in preview, with a context window of about one million tokens. Pricing is $4 per million input tokens, $20 per million output tokens and $0.20 per million cached input tokens. The API offers Standard mode only, which Meta's documentation says means your content is not used for training.
• Muse Wonder Full in Muse Code only. The stronger variant is not on the API. It is available only through Muse Code, Meta's terminal coding agent, on the Power Usage and Maximum Usage plans, which CNBC reports at $50 or $200 a month.
• The training default. Muse Code subscriptions default to Contributor mode, in which prompts, code, files and responses may be used to train future Meta models in exchange for more usage. It can be switched off in settings.

How far to trust the numbers

Several headline measures on Meta's model card are Meta's own benchmarks, and Meta says its runs of competitor models were "best-effort". Weigh the third-party rows, from Artificial Analysis, Vals AI and Datacurve, first. On price, Alexandr Wang told CNBC the API will be "meaningfully below" leading rivals, while TechCrunch puts list prices roughly level with them. Assume market-rate pricing until your own workload tests say otherwise.

On safety, Meta says it could not rule out the unmitigated model reaching its high-risk threshold for chemical and biological misuse, so it added safeguards. A fuller safety report is still to come.

What leaders should do next

1. Settle the data terms before the trial. Make Standard mode a written condition of any Muse Code approval, and check the setting on every seat.
2. Evaluate through the API. Its no-training terms are the default and usage is metered centrally.
3. Hold back broad agent permissions until the safety report lands, and require human confirmation for irreversible actions.
4. Check country availability before planning a pilot; Meta says it varies by country.

Sources and scope

Release details, availability, intended uses and safety results are from Meta's Muse Wonder announcement and model card, both dated 12 October 2026. API pricing, variant availability and the Contributor mode default are from Meta's developer documentation: overview, models and pricing and rate limits. All were checked on 13 October 2026. Subscription prices and Alexandr Wang's pricing remark are from CNBC, and the "first frontier model" framing and price comparison are from TechCrunch. Both are secondary reporting dated 12 October 2026. Benchmark results are Meta's reported figures and have not been independently reproduced by ByteNib. The recommendations are ByteNib editorial analysis.

Drafted with AI assistance and reviewed by the editor before publication.

Continue exploring: AI & Machine Learning analysis, the related implementation tutorial, and the structured learning path.