List your company

Multi-turn Tool-use on FTRL

42.41Solve Rate (R)PACT
20.736426.363231.9937.6168Jun 15, 2026

Evaluation results

MethodLinks
PACTBackbone=Qwen3-4B2026.0642.4134.6936.6
SFT→MatchTIRBackbone=Qwen3-4B2026.0637.1730.432.03
MatchTIRBackbone=Qwen3-4B2026.0636.130.631.6
GRPOBackbone=Qwen3-4B2026.0633.1231.830.84
CHORDBackbone=Qwen3-4B2026.0632.6929.8429.88
ToolRLBackbone=Qwen3-4B2026.0632.5928.9828.28
VanillaBackbone=Qwen3-4B2026.0632.1431.6628.6
FTRLBackbone=Qwen3-4B2026.0631.9833.2431.34
PACTBackbone=Qwen3-1.7B2026.0628.3326.9422.93
SFTBackbone=Qwen3-4B2026.0628.1929.3224.67
MatchTIRBackbone=Qwen3-1.7B2026.0624.1221.521.43
SFT→MatchTIRBackbone=Qwen3-1.7B2026.0623.3919.6419.72
ToolRLBackbone=Qwen3-1.7B2026.0622.7917.9317.99
FTRLBackbone=Qwen3-1.7B2026.0622.7822.0621.06
CHORDBackbone=Qwen3-1.7B2026.0622.6819.1418.5
SFTBackbone=Qwen3-1.7B2026.0622.3617.8917.66
VanillaBackbone=Qwen3-1.7B2026.0621.6620.7819.72
GRPOBackbone=Qwen3-1.7B2026.0621.5717.3718.13