Key Takeaways
- • Meta's Muse Wonder Full scored 56 on the Artificial Analysis Intelligence Index, tying Anthropic's Claude Sonnet 5.5, while the lighter Muse Wonder scored 52, one point behind OpenAI's GPT-6 Astra, according to the figures Meta and Artificial Analysis published October 12, 2026.
- • Coding results are split: Muse Wonder Full led Claude Opus 5.5 on SWE-Marathon (57.5% vs. 46.3%) but trailed on ProgramBench (59% vs. 70.5%).
- • Muse Wonder Full posted its clearest lead on MemoryBench at approximately 85.5% versus approximately 81.5% for Claude Opus 5.5, aligning with Meta's Active Memory feature.
- • Only the standard Muse Wonder is available through the Meta Model API at $4 per million input tokens and $20 per million output tokens; Muse Wonder Full is gated behind a Muse Code subscription.
- • No verified stock-price reaction or independently confirmed analyst note tied to the benchmark numbers had surfaced as of publication.
Summary
Meta released two models under the Muse Wonder name on October 12, 2026: Muse Wonder and the higher-scoring Muse Wonder Full. Both accept text, images, audio, and video as input, return text, and offer a context window of roughly one million tokens. On the Artificial Analysis Intelligence Index, a composite of ten evaluations, Muse Wonder Full scored 56, matching Claude Sonnet 5.5, and Muse Wonder scored 52, a point behind GPT-6 Astra. Artificial Analysis said Muse Wonder Full places in the top three on six of the ten evaluations and is particularly strong in agentic knowledge work and scientific reasoning.
Performance on coding benchmarks is split. On SWE-Marathon, which tests long-duration software engineering, Muse Wonder Full scored 57.5% against 46.3% for Claude Opus 5.5. On ProgramBench, which emphasizes discrete code generation and correctness, Claude Opus 5.5 led 70.5% to 59%. MemoryBench, which measures retention across sessions, gave Muse Wonder Full its clearest lead at approximately 85.5% versus approximately 81.5%, a result that aligns with the Active Memory feature Meta is promoting.
Access is structured in tiers. Developers can use Muse Wonder through the Meta Model API at $4 per million input tokens and $20 per million output tokens, with cached input at $0.20 per million tokens and rate limits of 3,000 requests per minute and 4 million tokens per minute. VentureBeat noted that third-party developers testing through the open API are not working with the model that produced Meta's top scores. Meta also said Muse Wonder will begin powering its consumer assistant, starting with subscribers, replacing Muse Spark as the default engine.
TechCrunch characterized Wonder as Meta's first genuine frontier model, and Meta CEO Mark Zuckerberg told reporters the company's models have reached the AI frontier, according to Yahoo Finance. Neither OpenAI nor Anthropic had issued a direct public response at the time of reporting. Meta's own chart still places Claude on top in some comparisons, and the coding split suggests the frontier claim is only partly supported by the benchmark data.
Performance on coding benchmarks is split. On SWE-Marathon, which tests long-duration software engineering, Muse Wonder Full scored 57.5% against 46.3% for Claude Opus 5.5. On ProgramBench, which emphasizes discrete code generation and correctness, Claude Opus 5.5 led 70.5% to 59%. MemoryBench, which measures retention across sessions, gave Muse Wonder Full its clearest lead at approximately 85.5% versus approximately 81.5%, a result that aligns with the Active Memory feature Meta is promoting.
Access is structured in tiers. Developers can use Muse Wonder through the Meta Model API at $4 per million input tokens and $20 per million output tokens, with cached input at $0.20 per million tokens and rate limits of 3,000 requests per minute and 4 million tokens per minute. VentureBeat noted that third-party developers testing through the open API are not working with the model that produced Meta's top scores. Meta also said Muse Wonder will begin powering its consumer assistant, starting with subscribers, replacing Muse Spark as the default engine.
TechCrunch characterized Wonder as Meta's first genuine frontier model, and Meta CEO Mark Zuckerberg told reporters the company's models have reached the AI frontier, according to Yahoo Finance. Neither OpenAI nor Anthropic had issued a direct public response at the time of reporting. Meta's own chart still places Claude on top in some comparisons, and the coding split suggests the frontier claim is only partly supported by the benchmark data.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.