Muse Wonder Full Scores 56 on AA Intelligence Index, Ranks Second Behind Claude Opus 5.5
Meta’s Muse Wonder Full, which launched on October 12 without a single published benchmark number, now has its first independent score: 56 on the Artificial Analysis Intelligence Index v4.3. That places it second overall behind Claude Opus 5.5 at 58, and ahead of Claude Fable 5.1 and OpenAI’s GPT-6 Astra.
The score gap from Meta’s previous frontier-tier model was not small. Muse Spark 1.3, released in early September, scored 48 on the same index. Muse Wonder Full adds 8 points in six weeks.
The AA Intelligence Index Table
Artificial Analysis’s Intelligence Index v4.3 aggregates model performance across a broad set of capability dimensions into a single comparable score. The current top of table:
| Rank | Model | AA Intelligence Index v4.3 |
|---|---|---|
| 1 | Claude Opus 5.5 | 58 |
| 2 | Muse Wonder Full | 56 |
| 3 | Claude Fable 5.1 | ~54 |
| 4 | GPT-6 Astra | ~52 |
Muse Wonder (the smaller sibling) scores 52 at max reasoning effort, landing at approximately the same position as GPT-6 Astra. Both Muse variants are a material step above Muse Spark 1.3 at 48.
First Data for a Benchmark-Free Launch
The October 12 launch post from Meta’s Superintelligence Lab contained no external benchmark figures. No SWE-bench, no Arena ELO, no LiveBench. The post described Muse Wonder Full as capable of “state-of-the-art performance on personal and professional agentic tasks” — a claim that required independent validation.
AA’s independent evaluation published the same day as launch. The Vals AI result followed: Muse Wonder Full debuted at fifth place on the Vals Index, solving 94 of 100 math problems and surpassing all GPT-6 variants tested on that benchmark.
Those two independent data points, arriving within 48 hours of launch, are what make this a meaningful benchmark result rather than a vendor claim. Meta’s prior models have had credibility gaps on independent evals; Wonder Full’s immediate appearance near the top of two major independent leaderboards is a different signal than a self-reported launch table.
The Cost Tradeoff
Muse Wonder Full costs approximately three times as much per API call as Muse Spark 1.3, according to Artificial Analysis pricing data. At the same time, Spark was positioned as a lightweight model optimized for latency; Wonder Full is a frontier reasoning model. Comparing per-call prices directly is misleading — they target different use cases.
The question for enterprise buyers: at 56 on the AA Intelligence Index, is Muse Wonder Full a credible alternative to Claude Opus 5.5 (58) at presumably similar or lower pricing? Meta has not published API pricing for Wonder Full yet. The answer depends on that number.
Meta’s Position in the Frontier Stack
Muse Wonder Full landing at #2 on AA’s Intelligence Index does not mean Meta has overtaken Anthropic’s full model lineup. Claude Opus 5.5 still leads. But it does establish that Meta has a model that:
- Scores above every OpenAI model on this specific index
- Scores above Google’s Gemini 4 Argon
- Sits close enough to the top that capability is no longer the argument against using it
That is a different competitive position than Meta occupied six months ago, when Muse Spark 1.3 scored 48 and sat outside the frontier tier proper.
The next data points to watch: SWE-bench Verified from vals.ai, Arena ELO, and pricing. The AA score is a strong opening indicator — it takes the launch claim from marketing language to a number that can be argued with.