Meta released Muse Spark 1.3 on September 3, 2026, saying it improves most on agentic tasks while still trailing top performers like Claude Fable 5.1 across most benchmarks. It is the company's fourth model in five months, following versions 1.1 in July and 1.2 in August.
Meta reported an index score of 62 points for the max tier and 61 for the xhigh tier, measured on the Intelligence Index. That compares with scores of 57 in August and 53 in July.
The jump comes down to how the index weights its tests, with GDPval-AA v2 counting for 20 percent, Terminal-Bench 2.1 for 16 percent, and τ³-Bench Banking for 14 percent.
Muse Spark 1.3 is built on Meta's internal architecture and targets agentic tasks such as tool operation in simulated banking scenarios. Availability begins now for the xhigh tier, initially for general users, while the more compute-heavy max tier runs as a limited partner preview.
"One index task costs $0.55," said Artificial Analysis. "No model scoring 59 points or higher is cheaper, and rivals at the same index level run between $0.94 and $1.23."
The announcement follows the release of Muse Spark 1.2, which ran at $0.40 per task. Meta did not say when the max variant will have a finalized price, and larger models and an open-weights release are on the way.
Source: thedecoder
