Meta released Muse Wonder and Muse Wonder Full on Oct. 12, 202612. They are the first frontier models from Meta Superintelligence Labs. Alexandr Wang has led that unit since 2025210. Both read a context window of 1 million tokens, the stretch of text a model holds in one conversation2. Claude Opus 5.5, GPT-6 Astra and Gemini 4 Argon all set that same ceiling678. Meta priced Wonder's API access at $4 per million input tokens and $20 per million output tokens13. Anthropic charges the identical $4 and $20 for Opus 5.56. Artificial Analysis, an independent evaluator, scores Wonder at 52 on its Intelligence Index4. That index is a single number built from ten tests of reasoning, coding and professional work. Opus 5.5 scores 58 on the same index, six points ahead, at the same price6. Wonder Full, the stronger of Meta's two new models, scores closer at 565. Meta sells it only inside a Muse Code subscription, apart from any per-token bill511.
The Price Matches Opus, to the Cent
Alexandr Wang set expectations before Wonder shipped. In August, Meta's AI chief picked price as Wonder's edge over raw capability12. This fall he told CNBC that Wonder's rate would land "meaningfully below" what OpenAI and Anthropic charge, Business Model Analyst reported12. Wonder's October price kept that promise against two of the three rivals in this story.
Wonder Charges Opus 5.5's Exact $4 and $20 Rate
Price per million input and output tokens, four frontier models, Oct. 17, 2026
Each pair of columns is one model: the shorter column is the input price and the taller one the output price, so two pairs the same height charge the same rate.
Sources: TechCrunch [1]; Artificial Analysis [6]; Artificial Analysis [7]; Artificial Analysis [8]. Chart by The AILately.com Desk.
The numbers behind this chart
| Item | Input price | Output price |
|---|---|---|
| Muse Wonder | $4 | $20 |
| Claude Opus 5.5 | $4 | $20 |
| GPT-6 Astra | $10 | $50 |
| Gemini 4 Argon | $2 | $10 |
GPT-6 Astra costs $10 per million input tokens and $50 per million output tokens7. Wonder's $4 and $20 run well under both lines. Gemini 4 Argon costs less than Wonder on both counts, at $2 and $108.
Against Opus 5.5, the model that actually tops the independent index, Wang's promise missed its mark. Anthropic charges $4 per million input tokens and $20 per million output tokens for Opus 5.56. Those are the same two numbers Meta charges for Wonder1. Wonder's cached input tokens cost $0.20 per million, a detail that rewards an app that repeats the same system prompt1. A startup running thousands of support tickets through a fixed instruction set would feel that discount every day.
Who Gets Wonder Full, and for How Much
Meta splits access to its new models two ways. The base Wonder model runs through the Meta Model API, billed per token like Opus 5.5 or Astra12. Wonder Full runs only inside Muse Code, Meta's subscription product211. Meta keeps it out of the API entirely11.
A Muse Code subscriber reaches Wonder Full for $50 a month on the Power Usage plan. Maximum Usage costs $200 a month instead11. Dr. Heinrich Vogel, who tracks AI pricing for the site Shattered, read the split plainly. "Meta is keeping it inside a walled garden of paid Muse Code plans for now," he wrote11. The Maximum Usage plan buys roughly 80 times the Power plan's usage allowance, by his count11. That gap points at who Meta expects to pay the top rate.
Meta's Scorecard Against the Independent Index
Artificial Analysis tests Wonder, Wonder Full and their rivals on its own, apart from Meta4. Its Intelligence Index combines ten tests into one score, and higher means stronger across the set. Opus 5.5 leads the group of four at 586. Wonder Full follows at 565. Astra and Argon tie at 5378. Wonder sits lowest at 524.
Wonder Scores Six Points Below Opus 5.5 at the Same Price
Price per million output tokens against Artificial Analysis Intelligence Index score, four frontier models; higher is better
Each dot is one model: its left-right position is the output price and its height is the Intelligence Index score, so a dot higher and to the left gives more for less.
Muse Wonder Full skips a per-token price, so this chart carries only the four models Meta and its rivals sell by the token.
Sources: Artificial Analysis [4]; Artificial Analysis [6]; Artificial Analysis [7]; Artificial Analysis [8]. Chart by The AILately.com Desk.
The numbers behind this chart
| Name | Price per million output tokens | Intelligence Index score |
|---|---|---|
| Muse Wonder | $20 | 52 |
| Claude Opus 5.5 | $20 | 58 |
| GPT-6 Astra | $50 | 53 |
| Gemini 4 Argon | $10 | 53 |
Meta's own benchmark card, published the same day as the launch, tells a less one-sided story3. Across 13 tests against Opus 5.5, Wonder wins some rows and loses others. It leads on ProgramBench, a coding test, 53.1 to 46.33. SWE-Marathon tells the same story. Wonder leads there too, 78.8 to 74.2, on a test that measures work across long sessions3. Opus answers back on DeepSWE 1.1, another coding test, winning 59.6 to 52.03. HealthBench Pro, a medical-reasoning test, goes the same way for Opus, 65.6 to 60.53. Astra beats both Wonder variants on MemoryBench, the row built to test the feature Meta markets hardest. It scores 43.1 against Wonder Full's 39.4 and Wonder's 34.33. Meta's own table leaves the cell for Opus 5.5 blank on that row3.
The Case For and the Case Against
Mark Zuckerberg made Meta's case the day Wonder shipped. "Muse Wonder is a major leap towards delivering personal superintelligence for everyone," he said, Decrypt reported10. His claim rests on reach, instead of rank. Wonder runs through an open API that any developer can bill against today.
Gabriel Anhaia made the opposite case from the spreadsheet. He is a senior software engineer at Ordio who writes about model releases for DEV Community. Opus 5.5 wins on most of Meta's own rows at that price, by his count. "The Wonder variant that beats Opus needs a $50-a-month Muse Code plan," he wrote9. His read turns on a plain fact. Meta's own table, instead of a rival's, supplies the rows he counted.
Read together, the two claims measure different things. Zuckerberg describes access, the thing Meta changed on Oct. 12. Anhaia describes value at the API's posted price, the thing Meta left short of a win.
What a Developer Buys
Wonder costs $4 and $20 per million tokens through the API1. That price buys the lesser of Meta's two new models. The stronger one, Wonder Full, needs a $50 or $200 Muse Code plan511. It stays out of any per-token bill11.
What the API price does buy is memory that persists across sessions. Kavya Kopparapu, a research scientist on the team that built it, said the group chased persistence above all1. "We put a tremendous amount of effort into making sure your model can remember what's important to you over days, weeks, months, and beyond." Meta calls the feature Active Memory. It carries context across conversations and pulls up old details only when a task actually needs them2.
The benchmark rows above show where that bet pays off and where it falls short. A developer running long coding sessions or multi-step agent tasks gets a reason to pick Wonder3. Memory only adds to that reason. Medical reasoning or short, isolated coding tasks tilt the other way. There, a buyer pays the same price Opus charges for a weaker result3. The memory feature is the one place Meta's pitch and its price might still line up. That holds when a workload actually rewards remembering.
By the numbers
- $4 and $20: Wonder's per-million price for input and output tokens, matching Opus 5.5's rate16.
- 52 to 58: Wonder's score against Opus 5.5's on Artificial Analysis's Intelligence Index46.
- 56: Wonder Full's score on the same index, reachable only through a paid plan5.
- $50 and $200: the monthly Muse Code plans that carry Wonder Full, Power Usage and Maximum Usage11.
- 1 million tokens: the context window Meta built into both models, matching Opus 5.5, Astra and Argon2678.
- 78.8 to 74.2: Wonder's lead over Opus 5.5 on SWE-Marathon, a long-session coding test, from Meta's own card3.
- DeepSWE 1.1: Opus 5.5 leads Wonder 59.6 to 52.0 on this coding benchmark, from the same card3.
- 80 times: the usage allowance Vogel reports the $200 Maximum plan buys over the $50 Power plan11.
What to watch
Watch whether Artificial Analysis scores Wonder Full on its own, once Meta opens it beyond Muse Code5. That would let a developer compare it to Opus 5.5 at an actual price. Meta's full evaluation report, still absent from the model card, should show scores Meta alone chose to run3. A rival lab's own harness would settle those rows a different way. Price is the lever a challenger usually pulls first when a score alone falls short of closing the sale. A Wonder discount from here would say the most about how this contest ends.
Sources
- Russell Brandom, "Meta releases its first frontier model, Muse Wonder," TechCrunch, Oct. 12, 2026, https://techcrunch.com/2026/10/12/meta-releases-its-first-frontier-model-muse-wonder/
- Meta, "Muse Wonder: A Frontier Model for Personal and Professional AI Agents," Meta AI Research, Oct. 12, 2026, https://research.meta.ai/blog/muse-wonder
- Meta Superintelligence Labs, "Put Your Ideas to Work With Muse Wonder and Muse Wonder Full," Meta for Developers, Oct. 12, 2026, https://dev.meta.ai/resources/blog/build-with-muse-wonder
- Artificial Analysis, "Muse Wonder (Max) - Intelligence, Performance & Price Analysis," Artificial Analysis, Oct. 17, 2026, https://artificialanalysis.ai/models/muse-wonder
- Artificial Analysis, "Muse Wonder Full (Max) - Intelligence, Performance & Price Analysis," Artificial Analysis, Oct. 17, 2026, https://artificialanalysis.ai/models/muse-wonder-full
- Artificial Analysis, "Claude Opus 5.5 - Intelligence, Performance & Price Analysis," Artificial Analysis, Oct. 17, 2026, https://artificialanalysis.ai/models/claude-opus-5-5
- Artificial Analysis, "GPT-6 Astra (Max) - Intelligence, Performance & Price Analysis," Artificial Analysis, Oct. 17, 2026, https://artificialanalysis.ai/models/gpt-6-astra
- Artificial Analysis, "Gemini 4 Argon (High) - Intelligence, Performance & Price Analysis," Artificial Analysis, Oct. 17, 2026, https://artificialanalysis.ai/models/gemini-4-argon
- Gabriel Anhaia, "Meta's Muse Wonder Costs Exactly What Opus 5.5 Costs. It Loses 9 of Meta's Own 14 Benchmarks," DEV Community, Oct. 12, 2026, https://dev.to/gabrielanhaia/metas-muse-wonder-costs-exactly-what-opus-55-costs-it-loses-9-of-metas-own-14-benchmarks-399a
- Jose Antonio Lanz, "Meta's New Muse Wonder Finally Puts Zuckerberg in the Frontier AI Race," Decrypt, Oct. 12, 2026, https://tech.yahoo.com/ai/meta-ai/articles/metas-muse-wonder-finally-puts-204604330.html
- Dr. Heinrich Vogel, "Meta Muse Wonder AI Pricing: $4-$200/Mo Explained," Shattered, Oct. 12, 2026, https://shattered.io/meta-muse-wonder-ai-pricing-2026/
- Flavia Moura, "Meta Promised Cheaper AI. It Matched Claude's Price to the Cent," Business Model Analyst, Oct. 13, 2026, https://businessmodelanalyst.com/meta-muse-wonder-api-price-claude-opus-match/