Meta released Muse Wonder and Muse Wonder Full on Monday, its first models built to compete at the frontier of artificial intelligence, and said in its announcement that they "deliver state-of-the-art performance on personal and professional agentic tasks." Its developer pages price Muse Wonder at $4 per million input tokens and $20 per million output tokens. Anthropic's price list has Claude Opus 5.5 at exactly the same figures. Meta's shares closed up 3.4 per cent.
Alexandr Wang, who runs Meta's AI effort, told CNBC on Monday that the model would still be priced "meaningfully below" leading models from OpenAI and Anthropic. Whether that is true depends on which models count as leading. The gap between those two statements is where the business story sits.
What Meta is selling
Muse Wonder is available through the Meta Model API and Muse Code, Meta's coding agent. Muse Wonder Full, which Meta describes as its most advanced model, is not sold by the token at all. It comes only with the Power Usage plan of Muse Code at $50 a month and the Maximum Usage plan at $200, according to Meta's Muse Code page. Consumers will meet the model later: Meta says Muse Wonder "will soon power experiences in Muse," its personal agent app, rolling out first to subscribers, whose plans cost $20 or $100 a month, CNBC reported.
The launch is the first frontier release since Wang restructured Meta's AI work in August 2025, TechCrunch reported. Meta says the models were trained at its Prometheus data center, which it calls gigawatt-scale. Wang told CNBC the model was trained so that agents meeting a setback would "try other approaches instead of stalling and then checking back in with the user at any time."
The price lists, side by side
Here is how the published prices compare per million tokens, input then output. Muse Wonder: $4 and $20. Anthropic's Opus 5.5: $4 and $20. Anthropic's Fable 5.1: $10 and $50. OpenAI's GPT-6 Astra: $10 and $50, on its API pricing page. OpenAI's GPT-6.1 Sol, launched on 29 September and covered here at the time: $2 and $10 for standard-length prompts.
By this publication's arithmetic, Muse Wonder is 60 per cent cheaper than Astra and Fable 5.1, level with Opus 5.5 and twice the price of Sol. "Meaningfully below" describes the first comparison, not the other two. Cached input is also level: $0.20 per million tokens at both Meta and Anthropic.
The subscription tiers line up the same way. Muse Code's $50 and $200 plans sit beside Anthropic's Max plan, which starts at $100 a month and includes Claude Code. None of this says which product buyers will choose. It says Meta has priced to match rather than to undercut.
What the benchmarks do and do not show
Every performance claim so far is Meta's or comes from charts Meta chose to publish. On the Artificial Analysis Intelligence Index, as charted in Meta's post and dated 12 October, Muse Wonder Full scores 56, behind Opus 5.5 at 58 and ahead of Fable 5.1 and GPT-6 Astra at 53; Muse Wonder scores 52. Meta's charts put Muse Wonder Full first on a legal benchmark built by Harvey and narrowly first on the Vals Finance Agent test, at 59.1 per cent against 58.9 for Fable 5.1 and 58.6 for Opus 5.5.
A gap of a few tenths of a point is not a lead a buyer can feel. Artificial Analysis lists both models as evaluated on Monday; broader independent testing has barely started, and Meta says a full safety report "will be available soon."
The bill behind the model
The model arrives on top of Meta's capital budget. In its second-quarter results on 29 July, Meta said it expects 2026 capital expenditure of $130 billion to $145 billion. It spent $31.08 billion in the quarter, about 51 per cent of its $60.80 billion of revenue by our arithmetic, and free cash flow was $784 million.
That is why the subscription line matters more than the token price. Reuters, in a report carried by Investing.com, said the Muse rollout could help Meta earn subscription revenue from its AI spending, despite privacy concerns. Some Muse users have complained the agent read personal messages without authorisation, which Meta disputes, CNBC reported. Meta did not give subscriber numbers on Monday.
The shares were up about 4 per cent in afternoon trading, Reuters reported, and closed at $742.96, up from $718.67 on Friday, after trading as high as $752.24, on about 24.7 million shares, roughly twice Friday's volume, according to Yahoo Finance data. The S&P 500 finished almost unchanged.
The rivals' numbers
The companies Meta is pricing against are private, so their figures are reported rather than audited. OpenAI raised $122 billion at an $852 billion valuation in March, Bloomberg reported, and has been reported to be seeking more at about $1.4 trillion, as this publication noted last week. Anthropic raised $65 billion at a $965 billion valuation in May, according to Orrick, which advised it.
Anthropic told investors its annualised revenue run rate reached $65 billion at the end of July, CNBC reported, with preliminary second-quarter revenue of $11.5 billion. OpenAI told investors its September run rate was approaching $50 billion, Reuters reported after the Financial Times, with second-quarter revenue of $6.7 billion. The two do not count alike: Anthropic includes sales made through cloud partners and OpenAI does not, according to Reuters. Meta's own revenue last quarter was more than five times Anthropic's reported figure, but $59.36 billion of the $60.80 billion, about 98 per cent, was advertising.
Whether Muse Wonder takes business from either rival is not something a launch day can show. That is our reading, not any company's account.
What to watch
The test is in Meta's next quarterly report: whether it discloses Muse or Muse Code subscription revenue, and whether capital spending holds inside the $145 billion ceiling. The other test is outside Meta's control. If independent evaluations confirm Meta's charts, matching Opus 5.5 on price is a contest. If they do not, it is just a price.
The release, availability, Active Memory and Active Focus, the Prometheus data center, the "fourteen months" overhaul and the "personal superintelligence" framing are from Meta's announcement of 12 October 2026. Muse Wonder's API prices, its 1 million token context window and the Muse Code plan prices and descriptions are from Meta's developer pages, read on 12 October. Benchmark figures are as charted in Meta's announcement; the Intelligence Index chart is credited there to Artificial Analysis and dated 12 October, and Artificial Analysis lists both models as evaluated that day. TechCrunch (Russell Brandom, 12 October) reported the August 2025 restructuring under Alexandr Wang and that the training took place in central Ohio. Wang's quotations, the "meaningfully below" pricing remark, the $20 and $100 Muse app plans and the privacy complaints are from CNBC's report of 12 October. Reuters (Jaspreet Singh, 12 October) reported the afternoon rise of about 4 per cent and the subscription-revenue point. Closing prices, the intraday high, volume and the S&P 500 close are from Yahoo Finance market data. Anthropic's prices are from its pricing page and OpenAI's from its API pricing page, both read on 12 October. Meta's capital expenditure, free cash flow, revenue, advertising revenue and guidance are from its second-quarter results release of 29 July 2026. OpenAI's $852 billion valuation is as reported by Bloomberg on 31 March; Anthropic's $965 billion valuation and $65 billion round are from Orrick's announcement of 29 May; Anthropic's $65 billion run rate and $11.5 billion preliminary second-quarter revenue are as reported by CNBC on 17 August; OpenAI's September run rate of nearly $50 billion, its $6.7 billion second-quarter revenue and the difference in how the two companies count cloud-partner sales are as reported by Reuters, following the Financial Times, in October. The percentage changes, price ratios, the advertising share and the capex share of revenue are this publication's arithmetic. The analysis is our own.
Topics businessMetaartificial intelligenceOpenAIAnthropicsubscriptions
