Meta unveils Muse Spark 1.3, but top results limited

Meta's announcement positions Muse Spark 1.3 as a major step forward in coding and agentic tasks, with Zuckerberg touting 'frontier performance almost too cheap to meter.' The gains over the 1.2 release from last month are described as significant, particularly for long-running agent tasks. This suggests Meta is focusing on practical utility for developers who need reliable AI for complex, multi-step workflows.

The caveat is that the strongest benchmark results come from the max reasoning mode, which isn't yet broadly accessible. This creates a gap between the model's advertised potential and what most developers can currently use. While Meta promises the mode will arrive shortly, the lack of an API provider for it, as noted by Artificial Analysis, means real-world deployment may still be some time away. For now, developers must weigh the accessible version's strong price-performance against the promise of even better results in the near future.