Meta has launched two variants of its new frontier model, Muse Wonder and Muse Wonder Full. The models demonstrate significant performance gains over their predecessor, Muse Spark 1.3, particularly in agentic evaluations and coding tasks.
On the Artificial Analysis Intelligence Index v4.3, Muse Wonder Full achieved 56 points, while Muse Wonder scored 52. The models feature new capabilities such as Active Memory for information retention and Active Focus for handling complex, multi-step tasks.
Coding performance saw notable improvements, with Muse Wonder Full reaching 68 points on the Coding Agents Index. Additionally, the model solved 94 out of 100 problems on the ProofBench mathematical test, compared to 58 for the previous version.
Despite these gains, factual knowledge remains a challenge, with scores of 30 and 25 on the AA-Omniscience test. However, hallucination rates for Muse Wonder Full dropped to 23 percent, down from 33 percent in the earlier model.
The cost for Muse Wonder is $4.52 per task, nearly triple the $1.60 cost of Muse Spark 1.3. Token pricing has increased to $4 per million input tokens and $20 for output tokens. Meta trained these models at its gigawatt-scale Prometheus data center.
