Meta’s Muse Spark 1.3 (Max) reaches No. 13 in Agent Arena, Arena says
Arena reports +4.2% net improvement across 8,700-plus real-world agentic sessions—a six-percentage-point gain over Muse Spark 1.2 (xHigh), which ranked No. 30.
TLDR
Arena’s September 11 results place Meta’s Muse Spark 1.3 (Max) 17 spots above Muse Spark 1.2 (xHigh), with net improvement rising from −1.8% to +4.2%. Arena reports the model reached No. 15 in both code and work; its chat score improved while its rank stayed at No. 17. In Arena’s signal-level breakdown, Confirmed Success rose from +1.5% to +8.4%, while Bash Recovery remained at +5.7%. Arena also reported improved praise-versus-complaint and steerability scores, and no tool-hallucination issues.
Combined views
5K
1 Source, first seen 19d ago