AutomationBench Benchmark Leaderboard | Phaseo

AutomationBench

Type: percentageAgents

Recorded Results

7

Average Score

22.03%

Score Range

13.50% - 30.80%

Leading Model (lowest score)

13.50% - Claude Sonnet 5

Scores Over Time

Individual benchmark scores plotted by date.

Models Using This Benchmark

Organisation	Model	Reported	Top Score	Info	Self Reported	Source
Anthropic	Claude Sonnet 5	30 Jun 2026	13.50%	Zapier leaderboard private held-out tasks; max effort	Yes	Source
Anthropic	Claude Fable 5	09 Jun 2026	17.40%	Higher of Mythos 5 / Fable 5 in Anthropic's launch table	Yes	Source
Anthropic	Claude Mythos 5	09 Jun 2026	17.40%	Higher of Mythos 5 / Fable 5 in Anthropic's launch table	Yes	Source
Anthropic	Claude Opus 5	24 Jul 2026	24%	Private held-out evaluation set; medium effort; $0.89 average cost per task	Yes	Source
DeepSeek	DeepSeek V4 Flash 0731	31 Jul 2026	25.10%	Public subset; DeepSeek-V4-Flash-0731	Yes	Source
Moonshot	Kimi K3	16 Jul 2026	30.80%	600-task public subset; otherwise follows the official GitHub setup.	Yes	Source