Mahal Systems τᶜ-Bench (TASTE)
Fits teams whose agent already does well on τ²-bench and who need harder cases on the same tools. Borrow the method: sample valid tool sequences, cluster them for coverage, then write the task around the sequence and tune its difficulty.
What it asks the agent to do
- complete a support task that requires an unusual combination of domain tool calls
- modify orders, bookings or telecom service under policy with a simulated user
- handle tasks generated from a tool-call sequence rather than a written scenario
What to borrow from it
case templatetaxonomy
Caveats
Harder extension of τ²-bench Verified (50 airline, 114 retail, 114 telecom tasks) rather than a new domain; the repo was not inspected here so license and runnability are unknown, and the tasks are LLM-generated and validated by LLM judges.
Details
| Workloads | customer support, telecom, retail, travel, tool calling |
|---|---|
| Interaction | multi turn simulated user |
| Environment | api sandbox |
| End state checked | Yes |
| Tools | the τ² bench airline, retail and telecom APIs |
| Grading | state diff, pass^k |
| Size | 278 |
| Data | synthetic |
| Personal data | none |
| License | unknown |
| Harness | none |
| Status | active |