Extended Benchmarks

This section introduces evaluation benchmarks that require additional dependency packages or special configurations. These benchmarks are typically designed for specific domains or tasks, providing more specialized evaluation capabilities.

Before using these benchmarks, please follow the instructions in each benchmark’s documentation to install the corresponding dependency packages and complete the necessary environment configuration.

• Terminal-Bench 2.0
• SWE-bench
• τ-bench
• τ²-bench
• BFCL-v3
• BFCL-v4
• Needle in a Haystack
• ToolBench
• LongBench-Write