List your company

Multi-turn Tool-use on FTRL, BFCL, and ToolHop Combined

36.9Average Success RatePACT
17.119222.254627.3932.5254Jun 15, 2026

Evaluation results

MethodLinks
PACTBackbone=Qwen3-4B2026.0636.9
SFT→MatchTIRBackbone=Qwen3-4B2026.0633.07
MatchTIRBackbone=Qwen3-4B2026.0632.93
FTRLBackbone=Qwen3-4B2026.0631.13
CHORDBackbone=Qwen3-4B2026.0629.35
GRPOBackbone=Qwen3-4B2026.0629.14
VanillaBackbone=Qwen3-4B2026.0628.31
ToolRLBackbone=Qwen3-4B2026.0628.25
SFTBackbone=Qwen3-4B2026.0625.12
PACTBackbone=Qwen3-1.7B2026.0624.9
MatchTIRBackbone=Qwen3-1.7B2026.0621.7
SFT→MatchTIRBackbone=Qwen3-1.7B2026.0620.78
FTRLBackbone=Qwen3-1.7B2026.0620.45
VanillaBackbone=Qwen3-1.7B2026.0619.06
CHORDBackbone=Qwen3-1.7B2026.0619
ToolRLBackbone=Qwen3-1.7B2026.0618.52
GRPOBackbone=Qwen3-1.7B2026.0618.08
SFTBackbone=Qwen3-1.7B2026.0617.88