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

SummaryAI generated

Meta released the latest AI model, Muse Spark 1.3. In third-party benchmarks, its maximum inference configuration achieved advanced performance, and there was significant improvement in cost-performance ratio compared to the previous version. The version currently available through Muse Code and Meta Model API uses xhigh inference settings; its maximum inference configuration is still under security testing, and no API providers have been confirmed yet. Although the maximum inference configuration scored slightly higher than the xhigh version in some tests, the difference was minimal, and in some tests, the results were even reversed. The Artificial Analysis index shows that although the xhigh version belongs to the advanced cluster, it still does not surpass Anthropic’s Claude Fable 5.1. Compared to the previous version, Muse Spark 1.3 performs better in coding and proxy tasks.

Related eventsRELATED EVENTS

Key entitiesKEY ENTITIES

Artificial AnalysisClaudeMark ZuckerbergMetaMuse Spark 1.3

Coverage · reports per dayLANGUAGE SPLIT

Entity relations

SignalsSIGNALS

Keyword heat

Meta1
Muse Spark 1.31
Mark Zuckerberg1
Artificial Analysis1
Claude1

All reports (1)SOURCES

↗

V VentureBeat AI en 2026-09-04 00:19

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 the latest AI model, Muse Spark 1.3, which achieved advanced performance in third-party benchmark tests with its maximum inference settings. Although this version has made significant improvements over version 1.2 in terms of long-running proxy tasks and price-performance ratio, only some configurations can be deployed immediately. Currently, the version using xhigh inference settings is available through Muse Code and Meta Model API; its maximum inference settings are still undergoing security testing, and no API providers have been identified yet. Although the maximum inference settings scored slightly higher than the xhigh version in tests like GDPval-AA and OSWorld, the difference between the two versions is minimal, or in some tests, it even reversed; the Artificial Analysis index shows that although the xhigh version is an advanced cluster, it still does not surpass Anthropic’s Claude Fable 5.1. Compared to last month’s version, Muse Spark 1.3 performs better in coding and proxy tasks, as shown by internal comparisons…