Skip to content
K = 0.731
Benchmark sources

τ³-Banking

Sierra / Professional work · Handles banking in a common tool environment, testing rule comprehension, customer communication and task completion.
Official benchmark
Across e/accAwaiting results
Evidence budget2%
Upstream data timeUnverified
Last successful syncAwaiting comparable results

What it measures and how

Compare only official standard banking, v1.0.1, alltools and the GPT-5.2 user simulator, using pass@1. Do not rank different tools, prompts, simulators, voice modes or custom systems together.

How this evidence is used

Counts toward ranking with 2% of the tools and office budget. The two environments sharing Mercor APEX’s allocation are adjusted to 4% in total. Banking does not establish coverage of law, e-commerce or every industry.

Limitations and attribution

Results belong to a fixed banking-task system and depend on simulated users and tool protocols. They are not general accuracy across all professions.

Data license: MIT · Official Sierra Research tau2-bench repository; retain source attribution.

Results are published by Sierra. Raw scores and the e/acc consensus index use different scales and cannot be added directly.