τ²-bench: Results as Published

tau2-bench tests a customer service agent in a conversation where both sides act: the agent and a simulated user each have tools that change a shared environment, so the agent must reason about the state and guide the user through steps the user alone can perform. The telecom domain (troubleshooting a phone line) is the dual-control setting; tasks are built from verifiable atomic components and a task passes when the final state matches the expected one.

Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.