Cua Driver Bench v2 (Linux, foreground)

Name: Cua Driver Bench v2 (Linux, foreground) results
Creator: Village

Cua's Cua Driver Bench v2 on Ubuntu 24.04 with XFCE and the xfwm4 window manager, with the desktop to the agent: ten desktop tasks (three multi-app workflows, three precision interactions and four interruptions) done only through a computer-use driver's tools, with no shell or file access. A trial passes when the task's checker confirms the apps' final state and no route out of the GUI was used; each task runs five seeded trials and the rate is passes over trials.

Models

Top score

84.0

Leader

Claude Sonnet 5.5

Updated

—

Resolution

—

Saturation

—

No signal or difficulty estimate. Held out of the fit by design.

6 dated models · 2 frontier modelsClosed weightsOpen weights

Show projectionFrontier only

Data: Scale AIatlas.village.dev

#	Model	Setting	Score	Source
1	Claude Sonnet 5.5	— · Cua Driver Bench on AgentEnv; Claude Code 2.1.280 with no --effort passed and its built-in tools limited to Skill and Read; trycua's cua-driver MCP tools alone; 45 turns and 360 seconds per trial	84.0	Cua Driver Bench on AgentEnv, linux foreground (2026-11-09 snapshot)
2	Claude Opus 5.5	— · Cua Driver Bench on AgentEnv; Claude Code 2.1.280 with no --effort passed and its built-in tools limited to Skill and Read; trycua's cua-driver MCP tools alone; 45 turns and 360 seconds per trial	76.0	Cua Driver Bench on AgentEnv, linux foreground (2026-11-09 snapshot)
3	Claude Haiku 5.5	— · Cua Driver Bench on AgentEnv; Claude Code 2.1.280 with no --effort passed and its built-in tools limited to Skill and Read; trycua's cua-driver MCP tools alone; 45 turns and 360 seconds per trial	68.0	Cua Driver Bench on AgentEnv, linux foreground (2026-11-09 snapshot)
4	GPT-6.1 Sol	medium · Cua Driver Bench on AgentEnv; Codex CLI 0.147.0 with every built-in Codex tool off; trycua's cua-driver MCP tools alone, whose screenshots Codex doesn't pass to the model; stopped after 45 tool calls or 360 seconds per trial	52.0	Cua Driver Bench on AgentEnv, linux foreground (2026-11-09 snapshot)
5	GPT-6.1 Terra	medium · Cua Driver Bench on AgentEnv; Codex CLI 0.147.0 with every built-in Codex tool off; trycua's cua-driver MCP tools alone, whose screenshots Codex doesn't pass to the model; stopped after 45 tool calls or 360 seconds per trial	52.0	Cua Driver Bench on AgentEnv, linux foreground (2026-11-09 snapshot)
6	GPT-6 Luna	medium · Cua Driver Bench on AgentEnv; Codex CLI 0.147.0 with every built-in Codex tool off; trycua's cua-driver MCP tools alone, whose screenshots Codex doesn't pass to the model; stopped after 45 tool calls or 360 seconds per trial	52.0	Cua Driver Bench on AgentEnv, linux foreground (2026-11-09 snapshot)