Skip to content

ApprenticeBench: end-to-end computer use, continual learning, and long-horizon agency on a real accounts-payable job

Tests whether a computer-use agent can learn a real accounts-payable job on the job, processing 100 vendor bills in a company ERP system with only the handbook, historical records, and mentor feedback a new hire would get.

18 models

Higher is better.

  1. 1Claude Fable 5.1Anthropic72
  2. 2GPT-6 AstraOpenAI68
  3. 3Claude Opus 5Anthropic36
  4. 4Claude Fable 5Anthropic34
  5. 5GPT-5.6 SolOpenAI26
  6. 6Gemini 3.8 FlashGoogle24
  7. 7GPT-5.5OpenAI20
  8. 8Muse Spark 1.3Meta19
  9. 9Kimi K3Moonshot AI18
  10. 10GPT-5.6 TerraOpenAI16
  11. 11Gemini 3.7 FlashGoogle16
  12. 12Claude Sonnet 5Anthropic16
  13. 13Grok 4.6xAI13
  14. 14GPT-5.4OpenAI11
  15. 15Claude Opus 4.7Anthropic7
  16. 16GPT-5.6 LunaOpenAI7
  17. 17Claude Opus 4.6Anthropic5
  18. 18Claude Sonnet 4.6Anthropic2
Higher is better. Bar colour: blue proprietary, orange open weights, green other.

About this benchmark

Metric
Cumulative success rate over 100 bills
Tasks
100 vendor bills processed in sequence inside a simulated construction company
Authors
NeoCognition (2026)

Scores are as reported by model developers or evaluators and may use different settings (for example reasoning effort or tools). Compare with care.