Computer-use training data

Navigate real apps.Complete complex work.

Train in working copies of enterprise software. Navigate screens, use multiple apps, and complete complex work with verified rewards.

Try 5 Samples
5,000tasks available
70+enterprise runtimes
8enterprise domains
Working software
  • screens
  • clicks + typing
  • multi-app work
Verified result
Runnable environments. Verified rewards.

The CUA catalog

Choose your enterprise domain.

Business workflows

Customers and teams

4enterprise domains
  • Marketing + retail
  • Customer service
  • Finance
  • Human resources

Navigate business applications through their interfaces. Complete work across customers, teams, and connected tools.

Operations workflows

Systems and services

4enterprise domains
  • IT service management
  • ML operations
  • Site reliability
  • Personal assistant

Operate the software behind technical workflows. Use real application interfaces to complete complex work across systems.

Choose your domains and the volume you need across 5,000 available CUA tasks.

Inside a CUA task: Beta-to-GA

A sound decision needs a visual story.

Turn nine rough Beta-to-GA slides into a board-ready recommendation. Choose the launch posture and build the visual argument.

9 rough slidesEvidence in workbooks
Visual argumentChoose the launch posture
Board-ready deckClear decision + next actions
152 atomic checksFacts, visuals + preservation

The task rewards a clear visual argument, beyond prose in boxes.Task workflow

Read the task and verification summary

Use the evidence in the workbooks to turn nine rough Beta-to-GA slides into a board-ready deck. Choose a launch posture and build a clear visual argument with a decision and next actions.

The sample contains 152 atomic checks covering facts, visuals, and preservation. Review the complete task package and reward verifier to evaluate fit against your own model.

Request the complete sample package ↗

Included with your tasks

Everything needed to run the work.

Inspect the complete task package in your samples.

Task specification
Runnable environment
Reward verifier
Customer-specific QA

Computer-use catalog evaluation

Hard tasks. Room to improve.

Frontier models pass 25–40% of the catalog’s tasks at pass@1. These results cover the broader computer-use catalog, evaluated in August 2026.

3models evaluated
pass@1success metric
Explore the full catalog ↗
40%best pass@1
Claude Fable 540%
GPT-5.6 Sol35%
Gemini 3.5 Flash25%

Your CUA mix. Your volume.

Choose your enterprise domains and the task volume you need.

1Review samples2Choose mix + volume3Get started now