τ³-Banking
Sierra / Professional work · Conducting banking business in a unified tool environment, testing rule understanding, customer communication and task completion.
What it measures, and how
Only the official standard banking, v1.0.1, alltools and GPT-5.2 user simulator are compared, using pass@1. Different tools, prompts, simulators, voice modes and custom systems are not mixed.
How this evidence is used
Officially scored; accounts for 2% of the tools and office budget, and the two environments shared with Mercor APEX are adjusted to 4% combined. Financial business does not mean law, e-commerce and all other industries are already covered.
Limits and data attribution
Scores belong to a fixed banking task system and depend on simulated users and tool protocols; this is not a general accuracy rate for all professions.
Data licence: MIT · official Sierra Research tau2-bench repository; retain source attribution.
Scores published by Sierra. Raw scores and the News consensus score use different scales and cannot be added directly.