CuaDriverBench AgentEnv - macOS / cua-driver 0.38.0 / Foreground: leaderboard

Publisher-reported historical macOS cua-driver 0.38.0 results from the Driver versions section. Exact five named native batches combine the complete five-run task cohorts used in that table:50 foreground and 40 background trials for each of four models, 360 trials total. Distinct preliminary/partial runs in the same parquet are retained as source evidence and are not substituted for the published historical cohort. Full pass requires task completion and GUI-only use; background additionally requires leaving the human user undisturbed. Native API identity, CLI harness/version and reported reasoning effort are retained. Score is 100 × passed / trials, without README rounding. This source remains separate from current macOS 0.47.10 and other OS/driver/condition boards. This board uses macOS, cua-driver 0.38.0, foreground condition; higher is better.

Metric: Full GUI-only pass rate (%). Source: huggingface.co. Saturation forecast: Around August 2027. 4 models tracked. Updated 14 Nov 2026.

Top models

#ModelScoreOverall rank
1Claude Opus 5.5 (Harness: claude-code v2.1.280; cua-driver v0.38.0; macOS)88
2Claude Sonnet 5.5 (Harness: claude-code v2.1.280; cua-driver v0.38.0; macOS)78
3Claude Haiku 5.5 (Harness: claude-code v2.1.280; cua-driver v0.38.0; macOS)76
4GPT-6.1 Sol60#10

No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.

Interactive version: theaggregate.ai/benchmark?slug=cuadriverbench-agentenv-macos-cua-driver-0-38-0-foreground · How It Works · Data refreshed daily, snapshot 2026-11-16.