τ³-Banking

Simulated banking support and troubleshooting.

97Public tasks

55.0%Highest score

The τ³-Banking leaderboard

97 banking-support scenarios test whether agents can retrieve policy and product information from a large document corpus, discover capabilities, and execute correct account changes.

Paper Code Sample task

Model

Score

Opus 5.5Max

55.0%

Fable 5.1Max

54.0%

Sonnet 5.5Max

54.0%

Gemini 3.8 FlashHigh

48.8%

GPT-6 AstraMax

46.7%

10%

20%

30%

40%

50%

60%

70%

80%

Pass@1

τ³-Banking

The τ³-Banking leaderboard

The latest on frontier AI performance, straight to your inbox.