Muse Wonder added to BenchGecko · Meta frontier model at $4/$20 per M

What landed

Muse Wonder is a frontier model from Meta, launched on 2026-10-12 on Meta Model API and Muse Code. The API model id is muse-wonder-1.0, labelled preview. BenchGecko lists the row as muse-wonder-1-0.

Meta released a second tier at the same time. Muse Wonder Full has no API and no per-token price. It is available only inside Muse Code subscriptions.

Pricing and context

From Meta's model page:

• Input · $4 per M tokens
• Cached input · $0.20 per M tokens
• Output · $20 per M tokens
• Context window · 1M tokens

Muse Wonder Full is included with two Muse Code plans:

• Power Usage · $50 per month, 20x the Everyday Usage limit
• Maximum Usage · $200 per month, 80x the Everyday Usage limit

BenchGecko stores the API rates for muse-wonder-1-0. The context field stays empty until Meta publishes an exact token count.

Capabilities Meta describes

• Inputs · text, images, screenshots, charts, documents and video. Meta says audio support needs confirmation on each release surface.
• Output · text
• Agent features · active memory across conversations, active focus on long tasks, sub-agent delegation, parallel tool calls and error recovery
• Computer use · operating browsers, desktop and mobile apps
• Access · an OpenAI SDK compatible endpoint on Meta Model API

Benchmarks

Meta's launch page compares Muse Wonder and Muse Wonder Full with other frontier models at max effort and says Muse Wonder Full leads DeepSWE and SWE-Marathon in third-party runs. BenchGecko has not added those cells yet because the published table does not map cleanly to rows. No BenchGecko test results exist for this model yet.

Sources

• Muse Wonder model page
• Build with Muse Wonder (Meta developer blog)
• BenchGecko model page: /model/muse-wonder-1-0

Source: BenchGecko