Muse Spark 1.3 (Medium): benchmark results

Provider: Meta. Released 2026-09-02. Access: API.

Aggregate Rating 1781 ± 22, rank #195 of 2274 rated models, from 23 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
Coarena - Task Completion96.2Completion Rate (%)100
ObviousBench98.61Answer pass³ (%)83.2
DuelLab GameBench 2 - Reasoning Variants64.44GameBench2 normalized score (0–100)76.5
Maze-Bench40.75Mean Score62.5
Thrive Helpdesk - Difficult Tickets40.38Correct ticket resolution (%)41.1
cua-speedrun - OSWorld (50 Tasks)77.54Mean verifier score (%), partial credit over the task set40
Coarena987.2Elo (Bradley-Terry)36
Argo-Bench - Solved9.52Share of the 210 tasks solved (%), a task counting as solved31
Bug Hunt Bench - VS Code Extension5.3Planted Bugs Fixed (out of 45)24.4
Web Clone Bench v1 - DocuSign0.4Full score on the DocuSign clone (%)24.4
Web Clone Bench v1 - Gmail8.7Full score on the Gmail clone (%)23.1
Web Clone Bench v1 - Slack4.9Full score on the Slack clone (%)23.1

Interactive version: theaggregate.ai/model?slug=muse-spark-1-3-medium · How It Works · Data refreshed daily, snapshot 2026-11-18.