Tool Use research benchmarks

Benchmarks

Dataset NameSOTA methodMetricTrendResultsLast Updated

τ-bench

FinTool-Qwen3-14B

38.3τ-bench Score

Mar 26, 2026

StableToolBench trace-free (test)

Trace-Free+

6.8F1 Score (Impr Pts)

Feb 26, 2026

MCP-Mark

Seed2.0 Lite

46.7Score

Jul 2, 2026

tau2-Telecom

MOPD

95.3Accuracy

Jun 30, 2026

MCPBench (All)

SKILLGEN

82.5Accuracy

May 13, 2026

DexSim2Real Six-Task Benchmark

DexSim2Real

77.2Sim Success Rate

May 8, 2026

Private Terminal Tool Calling

GRPO

88Accuracy

Mar 5, 2026

DexManip Tool (test)

0.6Computation Time (sec)

Feb 26, 2026

tau2-Bench retail

Seed2.0 Lite

90.9Score

Jul 2, 2026

Terminal-Bench 2.0

Otrue

25.39Cost (cents)

Jun 26, 2026

ToolRL prompts (hold-out)

Model A (ToolRL)

19Pre-Reconstruction Tool Correctness

Jun 26, 2026

Medical queries on-topic (test)

SFT

96Average Appearance Score

Jun 8, 2026

General queries off-topic (test)

SFT

52Average Appearance Score

Jun 8, 2026