Meta says Muse Spark 1.3 has frontier performance — but its best results come from a model developers can’t broadly use yet

Meta released Muse Spark 1.3, which surpasses version 1.2 on long-running agent tasks and attains frontier scores such as 1,754 Elo on GDPval-AA v2 and 66.9 on OSWorld 2.0. The top “max” configuration is still in safety testing and unavailable, while the public “xhigh” variant matches frontier-cluster performance comparable to GPT-5.6 Sol and Claude Opus 5; Anthropic models remain the overall leaders. Pricing stays at $1.25 / M input tokens, $4.25 / M output tokens, and $0.15 / M “fuel” tokens.