AI Agent Evaluations

Evaluations of AI coding agents challenges for Nitro

Last updated September 4, 2026.

ModelAgentAvg DurationAvg List CostSuccess RateSuccess Rate with AGENTS.mdFirst-Try Rate
GPT 5.6 Sol (xhigh)Codex1m 31s$0.126100.0% (41/41)100.0% (±0.0)100.0%
Claude Opus 5Claude Code2m 20s$0.682100.0% (41/41)100.0% (±0.0)100.0%
Claude Fable 5.1Claude Code2m 14s$0.578100.0% (41/41)100.0% (±0.0)100.0%
Gemini 3.7 FlashOpenCode5m 35s$0.28497.6% (40/41)100.0% (+2.4)97.6%
Claude Sonnet 5Claude Code1m 49s$0.30097.6% (40/41)100.0% (+2.4)97.6%
Kimi K3OpenCode3m 29s$0.27297.6% (40/41)100.0% (+2.4)90.2%
GLM 5.3OpenCode4m 17s$0.12797.6% (40/41)100.0% (+2.4)90.2%
Cursor Composer 2.5Cursor1m 18s$0.10797.4% (38/39)100.0% (+2.6)97.4%
Kimi K2.7 CodeOpenCode2m 22s$0.13295.1% (39/41)97.6% (+2.4)87.8%
Gemini 3.8 FlashOpenCode4m 54s$0.39992.7% (38/41)97.6% (+4.9)70.7%
Gemini 3.1 Pro PreviewOpenCode4m 05s$0.37790.2% (37/41)100.0% (+9.8)85.4%
MiniMax M3OpenCode2m 05s$0.04990.2% (37/41)95.1% (+4.9)73.2%
Claude Haiku 4.5Claude Code1m 09s$0.19243.9% (18/41)78.0% (+34.1)29.3%