Muse Spark 1.2: benchmark results

Meta's Muse Spark 1.2 model, a coding-focused update to Muse Spark 1.1 released alongside the Muse Code terminal agent; API-only, with no disclosed parameter count. Provider: Meta. Released 2026-08-05. Access: API.

Unified ELO 1941 ± 9, rank #37 of 2650 rated models, from 106 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
Vals AI Harvey Legal Agent Bench25.42Accuracy (%)100
Dnotitia Korean LLM Leaderboard100Mean Item Score (%)99.8
Design Arena (Data Viz)1365Elo99.4
Vals AI Finance Agent v260.6Accuracy (%)98.3
AI for Education SEND88.07Accuracy (%)98.1
Appwrite Arena (With Skills)97.8Overall Score (%)97.8
AI for Education Pedagogy - Maths94.44Accuracy (%)97.5
BenchmarkList ECI148.45Capability Index (ECI)96.4
Design Arena (Website)1321Elo96.3
Design Arena (ASCII Art)1323Elo95.4
Tinybird AI SQL Benchmark - Exactness57.22Result exactness vs human reference queries (0-100)95.1
ChessBench - Coherence86Coherence (%)95

Interactive version: theaggregate.ai/model?slug=muse-spark-1-2 · How It Works · Data refreshed daily, snapshot 2026-09-20.