3-Line TL;DR
- Meta released Muse Wonder and Muse Wonder Full on October 12, and Wonder Full scores 56 on the Artificial Analysis index, below Claude Opus 5.5 (58) but above GPT-6 Astra (53)
- Counting each lab's best model, Meta is now second on that index, but Wonder Full only comes with a Muse Code plan from $50 a month, and Meta's own table still shows clear losses
- I use Claude and am staying there for now, but Wonder Full looks worth a one-month test for long coding work
Wonder, then Wonder Full: say it out loud
- Meta named the pair Muse Wonder and Muse Wonder Full
- Read the bigger one aloud and you get "wonderful," a very developer kind of joke
- Wonder is the regular model, and Wonder Full is the heavier variant Meta calls its most capable (model card)
- Both take text, image, video and audio, answer in text and have a 1M-token context window
- Meta's pitch is Active Memory, which carries context across conversations, and Active Focus, which keeps working through blockers
56 points puts Meta between Claude and OpenAI
Source: Artificial Analysis article and leaderboard, checked October 13, 2026
- The index rolls ten outside evaluations, from coding to science questions, into one score
- Wonder Full gets 56, tied with Claude Sonnet 5.5 and two points behind Opus 5.5
- OpenAI's best here, GPT-6 Astra, gets 53, the same as Gemini 4 Argon and Claude Fable 5.1
- Taking each lab's top model, that is Anthropic first and Meta second
- Artificial Analysis links much of the jump to fewer hallucinations: 23% for Wonder Full, down from 33% for Muse Spark 1.3
- Two weeks ago I called Meta the leader of the second tier at 48 points, and that label aged fast
Llama scores 10 on the same index
Source: Artificial Analysis Llama 4 Maverick page (marked as an estimate) and Muse Wonder article, checked October 13, 2026
- Meta used to be known for Llama, open-weight models anyone could download and run
- Its last big one, Llama 4 Maverick from April 2025, sits at an estimated 10 on today's index
- Meta rebuilt its AI effort from August 2025, and Muse Spark and now Wonder came out of it
- The trade-off is openness: Muse Wonder is used through Meta's API and subscriptions, with no downloadable weights
- Going from open-model contributor to the second-best lab on this chart in about 14 months is a big turnaround
Meta's own table: long coding wins, some clear misses
Source: Muse Wonder model card, benchmark table, max effort; Meta-selected comparison
- This is the maker's own scorecard, so the losses it admits are the useful part
- Wonder Full leads SWE-Marathon, a test of very long coding projects, 57.5 to Opus 5.5's 46.3
- That row mixes agent tools (Codex, Claude Code) and runs Wonder Full at xhigh, so it is not a like-for-like test
- The MemoryBench lead, 85.5 to 81.5, comes from a benchmark Meta built itself
- Opus 5.5 stays ahead on Terminal-Bench 4.0, HealthBench Pro and ProgramBench, by 11.5 points on the last one
- Meta also lists 3D and spatial reasoning as a known weak spot
Wonder costs the same as Opus, Full needs a plan
Source: Meta model pricing, Anthropic pricing, Yahoo Tech; USD before tax, checked October 13, 2026
| Muse Wonder | Muse Wonder Full | Claude Opus 5.5 | |
|---|---|---|---|
| AA index | 52 | 56 | 58 |
| API price per 1M tokens (in / out) | $4 / $20 | No API price | $4 / $20 |
| AA cost per index task | $4.52 | n/a | $5.98 |
| Where to get it | Meta Model API, Muse Code | Muse Code $50 or $200 a month | Claude API and plans |
- Wonder lists at exactly Opus 5.5's token price, while scoring six points lower on the index
- It used fewer tokens in Artificial Analysis's run, so a full index task cost $4.52 against $5.98 for Opus
- Wonder Full has no per-token price; it comes with the Muse Code Power plan at $50 a month or Maximum at $200
- So the model with the "wonderful" name is the one developers cannot call directly yet
Muse the assistant drew people in first
- Meta's Muse personal-agent app launched on September 8 and hit No. 1 on Apple's US free iPhone chart on September 18 (Yahoo Tech)
- The same report cites Sensor Tower at more than 730,000 US downloads in the first ten days
- On Google Play it was only No. 338 in Productivity, and downloads do not show how many people kept using it
- Meta says Wonder will power Muse soon, starting with subscribers
- A much stronger model landing in an app people already installed is a smart order of moves, and Meta seems to be doing a lot right
I'm on Claude, but I'd test Wonder Full
- I use Claude today, and this is the first Meta model I actually want to try
- My test would be one $50 Muse Code month on long coding tasks, the area where Meta's own table looks strongest
- I have not used Wonder or Wonder Full yet, so everything above is published results
Community Reactions
- A Hacker News commenter said second place on Artificial Analysis, behind only Opus 5.5, is "really really not bad" (Hacker News)
- Another pointed out that the scores on Meta's marketing page belong to Wonder Full, which only comes with higher-tier plans, while the API price is for Wonder (Hacker News)
- One user's image-to-HTML test came out much worse than Opus 5.5, closer to Sonnet or Haiku, and they suspect the vision side is not mature yet (Hacker News)
- A lower-tier Muse Code subscriber said two fairly small Wonder prompts used 4% of their weekly quota, so they switched back to Muse Spark 1.3 (Hacker News)
Q&A (Field Notes)
- Q. Is Wonder Full better than Claude Opus 5.5?
- Not on the Artificial Analysis index (56 vs 58); Meta's own table shows wins on long coding and memory and losses on terminal, health and ProgramBench tasks
- Q. Can I download and run it like Llama?
- No; Meta offers Wonder through its API and Muse Code, and Wonder Full only through Muse Code plans
- Q. Is Muse Wonder a cheaper Opus?
- The token price is the same; Artificial Analysis measured a lower cost per task because Wonder used fewer tokens, but it also scored lower
- Q. Can I use the Muse app outside the US?
- The launch report covers the US on iOS, Android, WhatsApp and the web; I did not find confirmed availability for other markets