# Meta's Muse Wonder Full beats Claude Opus 5.5 on 8 of 14 Meta-published benchmarks, but only the weaker variant is sold through the API

The base Muse Wonder that developers can call wins five of those 14 comparisons, costs the same as Opus 5.5 at $4 per million input tokens and $20 per million output tokens, and Meta has released no weights.

- Published: 2026-10-13T06:13:53.502Z
- Canonical: https://polylog.news/ai/2026-10-13/meta-s-muse-wonder-full-beats-claude-opus-5-5-on-8-of-14-met
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [Polylog editors](https://polylog.news), [Meta AI](https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/), [VentureBeat](https://venturebeat.com/technology/metas-new-muse-wonder-and-muse-wonder-full-models-achieve-frontier-performance-but-only-the-weaker-one-is-available-via-api), [Pasquale Pillitteri](https://pasqualepillitteri.it/en/news/23176/muse-wonder-launches-opus-5-5-prices-trails-benchmarks), [LeoDex News](https://news.leodex.io/news/meta-s-muse-wonder-full-ranks-second-on-artificial-analysis-index-behind-claude-), [Forkast](https://forkast.news/meta-ships-muse-wonder-its-first-frontier-model-and-leaves-the-research-blog-empty/)

Meta Superintelligence Labs (MSL) released Muse Wonder, a two-tier model line for coding and long-running agent tasks. Both tiers have a 1,048,576-token context window, [according to VentureBeat](https://venturebeat.com/technology/metas-new-muse-wonder-and-muse-wonder-full-models-achieve-frontier-performance-but-only-the-weaker-one-is-available-via-api). The Russian-language channel [AI ML Big Data](https://t.me/ai_machinelearning_big_data/11115) described the launch as Meta's return to the frontier. [AI Post](https://t.me/aipost/8459) said the top variant beats Claude Opus 5.5 on 8 of 14 published benchmarks and leads GPT-6 Astra and Claude Fable 5.1 on long-running coding, legal work and business tasks. Those comparisons come from Meta's own tables.

A row-by-row reading of Meta's numbers [by Pasquale Pillitteri](https://pasqualepillitteri.it/en/news/23176/muse-wonder-launches-opus-5-5-prices-trails-benchmarks) found that Muse Wonder Full wins eight rows and loses six against Opus 5.5. The base Wonder model wins five and loses nine. Full's largest leads are on legal and agentic coding: 13.1 versus 4.2 on Harvey LAB-AA, 79.0 versus 74.2 on DeepSWE 1.1, and 57.5 versus 46.3 on SWE-Marathon. Opus 5.5 leads on HealthBench Pro (65.6 versus 62.3), on OSWorld computer use (81.8 versus 78.4) and on ProgramBench (70.5 versus 59.0), where the gap is widest.

Independent measurement places Full behind Anthropic's flagship model. On the Artificial Analysis Intelligence Index, Muse Wonder Full scores 56, tied with Claude Sonnet 5.5 and behind Opus 5.5 at 58, [LeoDex reported](https://news.leodex.io/news/meta-s-muse-wonder-full-ranks-second-on-artificial-analysis-index-behind-claude-). The 8-of-14 headline therefore depends on which benchmarks Meta chose to publish. The legal and long-horizon coding evaluations where Full leads most are also the evaluations with the fewest independent reproductions.

The distribution terms matter as much as the scores. Developers can call base Muse Wonder through the Meta Model API, the commercial endpoint Meta opened with [Muse Spark](https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/), at the same $4 input and $20 output per million tokens as Opus 5.5. Muse Wonder Full is not in the API. It runs only inside Meta's Muse Code assistant, on the $50 per month Power Usage tier or the $200 per month Maximum Usage tier, [VentureBeat reported](https://venturebeat.com/technology/metas-new-muse-wonder-and-muse-wonder-full-models-achieve-frontier-performance-but-only-the-weaker-one-is-available-via-api). Neither tier has downloadable weights. [Forkast reported](https://forkast.news/meta-ships-muse-wonder-its-first-frontier-model-and-leaves-the-research-blog-empty/) that Meta published no research write-up with the release, so parameter count, architecture and training compute remain undisclosed.

## What this means

Meta is competing for coding revenue through its own subscription product, not through an open model or a lower price. That exposes Anthropic's Claude Code and the third-party coding tools that resell Opus tokens. Developers on the API get a model that trails Opus on most of Meta's own rows and costs the same, so the API offer carries no clear price or performance advantage. Many observers are describing Muse Wonder as Meta "back at the frontier." The more precise picture is narrower: the model sits near the frontier on an independent index and leads only on the long-horizon evaluations Meta selected for publication. The closest precedent is Meta's move from open Llama weights to the closed Muse Spark API in April 2026, which ended Meta's role as the main US supplier of open frontier weights and left that position to Chinese labs. The likelier outcome is that within weeks to a few months Meta either cuts the Wonder API price below Opus or exposes Full through the API. Absent one of those moves, API adoption is likely to stay thin. This forecast depends on Meta wanting API share and not only Muse Code subscribers. It would be wrong if base Wonder gains meaningful routing share on aggregators such as OpenRouter at parity pricing. For investors, the release supports Meta's argument that its capital spending produces competitive models. It does not yet show a revenue line comparable to Anthropic's.

## What to watch

- Whether Artificial Analysis or other independent groups reproduce Meta's SWE-Marathon and DeepSWE leads. Those two rows carry most of the claim that Meta is ahead.
- Whether Meta adds Muse Wonder Full to the Meta Model API or reprices base Wonder below Opus 5.5. Either move would show Meta is competing for developer share and not only for Muse Code subscribers.
- Whether Meta publishes a model card or technical report with parameter count and training compute. Without one, outsiders cannot judge efficiency claims or assess safety.