Meta Matched Claude Opus 5.5's API Price, Giving 2027 AI Budgets a Fourth Bid
Meta put Muse Wonder on its Model API at $4 and $20 per million tokens, Claude Opus 5.5's exact list rate. In July, Muse Spark 1.1 launched there at $1.25 and $4.25.
TL;DR
- Meta launched Muse Wonder on October 12 at $4 input and $20 output per million tokens on its Model API.
- That matches Claude Opus 5.5's list price, so Meta adds a fourth frontier bid to 2027 budgets rather than a discount.
- If Meta's API leaves public preview at $4 and $20, Muse Wonder competes on quality, where Artificial Analysis scores it 52 to Opus's 58.
On October 12, Meta introduced Muse Wonder and Muse Wonder Full, two models it says bring it to the frontier on what it calls personal and professional agentic work. Muse Wonder went on sale the same day through the Meta Model API. On Meta's own Model API price table, it costs $4 per million input tokens, $0.20 for cached input and $20 per million output tokens.
Muse Wonder Full, the stronger of the two, isn't on the API at all. Meta sells it only inside its Muse Code subscription.
Meta's research post backs that with comparison tables, run on an evaluation suite Meta picked. GTM teams can leave the benchmark argument alone. For anyone writing a 2027 AI budget, a fourth lab now sells a model it pitches as frontier-class by the token, at the incumbent's rate.
Meta priced Muse Wonder at Claude Opus 5.5's rate, to the cent
Anthropic's published pricing for Claude Opus 5.5 is $4 per million input tokens, $20 per million output tokens and $0.20 for a cache read. Muse Wonder's line on Meta's table matches all three numbers.
The match marks a change of posture for Meta. When it opened the Model API in public preview on July 9 with Muse Spark 1.1, the rate was $1.25 in and $4.25 out, about a fifth of Muse Wonder's output price. Muse Spark still sells at that rate.
Independent scoring doesn't make Muse Wonder the better buy at that price. Artificial Analysis, which runs its own index rather than Meta's suite, puts Muse Wonder at 52 and labels it "somewhat expensive" next to models of similar price. It scores Claude Opus 5.5 at 58 for the same $4 and $20. Muse Wonder Full scores 56, but that's the model a developer can't call through the API.
Meta has put its best API model up to be bid against Anthropic on equal terms, with no discount to buy share. For a buyer, that still helps.
A second credible quote at the same list price is what a procurement team needs to ask its current lab for committed-spend discounts, and Meta made switching cheap to test. Its API page says teams can point existing OpenAI SDK code at it. OpenRouter also lists Muse Wonder at the same rates, so a team already routing models there can add it as one more option.
Two weeks earlier, Meta hired MongoDB's CEO, CJ Desai, to run a new Meta Enterprise Platform, which we read as Meta buying the enterprise half of its AI story. Muse Wonder on the API looks like the product half, something for Desai's team to sell into budgets that now renew with OpenAI, Anthropic or Google by default.
Meta's API is still in preview, and Opus scores higher at the same price
The API is still labeled public preview on Meta's own page, and beyond a Contact Sales link, nothing on it describes enterprise service levels, regional data residency or volume pricing. Those are the items a security review and a procurement team ask for first.
Meta's announcement says a full report with more safety evaluations will be available soon, so a vendor-risk questionnaire filed today would come back with blanks. Meta's table also lists cheaper "contributor" versions of Muse Spark, and their traffic is "used to improve our products." Your legal team will want to read that data-use choice before anyone picks a model ID.
There's also the plain cost math. At identical prices, a model that scores six points lower on an independent index has to win on your own tasks to justify the switch. Meta's benchmark tables can't tell you whether it does.
Meta moved from a fifth of the price to parity. The index still favors Opus
Meta's Model API opened on July 9 with Muse Spark 1.1 at $1.25 in and $4.25 out. Three months later, Muse Wonder lists at $4 and $20, the rate Anthropic charges for Claude Opus 5.5. That puts a fourth quote in front of every 2027 budget that now renews with OpenAI, Anthropic or Google by default.
If the API leaves public preview with $4 and $20 still on Meta's price table, the fourth bid competes on quality at Opus's price. There, Artificial Analysis scores Muse Wonder at 52 and Claude Opus 5.5 at 58.
Meta has put its best API model up to be bid against Anthropic on equal terms, with no discount to buy share.
The next things to watch are Muse Wonder's price when Meta's Model API leaves public preview and the fuller safety report Meta says is coming soon.
