What it means that Meta launched Muse Wonder to take on OpenAI and Anthropic — October 13, 2026
· CompaniesAutomation
Radar Flash Edition: Meta joins the front row with Muse Wonder, priced like Claude Opus 5.5. What to test and which mode to turn off before giving them your data.
Flash edition. In this morning's Radar, we suggested signing AI contracts for one year with a price review, not three at a locked rate. Since Monday afternoon there's one more reason: Meta has joined the AI frontier, and its API costs the same as Claude Opus 5.5.
What happened
On Monday, Meta introduced Muse Wonder and Muse Wonder Full, its most capable models, built for agents that run for hours, remember context and check in before anything important (Source). Muse Wonder is available through the API at $4 per million input tokens and $20 per million output tokens; Muse Wonder Full, the more powerful of the two, is only available inside Muse Code, Meta's coding assistant, on $50 and $200 monthly plans (Source). On Artificial Analysis' independent Intelligence Index, Wonder Full scores 56, above OpenAI's GPT-6 Astra (53) and below Anthropic's Claude Opus 5.5 (58), which costs the same per token as Wonder; Wonder sits at 52 (Source).
Why it matters
Meta no longer gives its best work away: Llama models could be downloaded and run on your own servers; Wonder can't. And Meta needs to get paid: on the same Monday, Zuckerberg said on a podcast that in late 2025 he was ready to write a check for $100 billion or $200 billion before knowing whether the model would work, with capital spending guided at $130 billion to $145 billion this year (Source). For buyers, counting Google, that makes four labs at the frontier. But the same price per token doesn't mean the same price per task: Artificial Analysis puts the cost of its average task at $4.52 with Wonder, $5.98 with Opus 5.5 and $3.26 with GPT-6 Astra (Source).
What it means for your business
- Run a switch test this week. Meta's API works with OpenAI- and Anthropic-compatible developer tools: put 20 of your real tasks through Wonder and compare cost per completed task, not per token.
- Turn off Contributor mode before you try Wonder Full. Muse Code subscriptions have it on by default, which lets Meta use your prompts, code, files and the model's responses to train its models. Muse Wonder on the API runs in Standard mode, which doesn't train on your content (Source).
- Don't sign up for exclusivity. With four providers at the top, the price review in your contract finally has leverage.
On the AI First ladder, being able to swap models in an afternoon using your own tests is part of being INTEGRATED; if you can't, you're locked in to one vendor.
Frequently asked questions
What is the difference between Muse Wonder and Muse Wonder Full?
Wonder is sold through the API; Full, the more powerful model, is only available in Muse Code, on $50 and $200 monthly plans. Neither can be downloaded.
Is it better than ChatGPT or Claude?
In Meta's own charts, Wonder Full leads on long-horizon coding and business workflows, and trails on health and terminal tasks; on the independent index it ranks behind Claude Opus 5.5.
Does Meta train on my company's data?
Not through the API in Standard mode. In Muse Code, Contributor mode is on by default and does allow it; you can switch it off in settings.