List your company

Tool-use alignment on FA-Bench 500 queries

91TCRToolRLA
60.8468.6776.584.33Mar 2, 2026

Evaluation results

MethodLinks
ToolRLALatency=1.6s2026.0391142.8960.8
ReAct + GRPO (coarse)Latency=1.9s2026.0382213.1746.5
ReAct + GRPO (additive)Latency=1.8s2026.0380223756.2
ReAct + PPO (binary)Latency=2.1s2026.0376263.4708
ReAct + SFT onlyLatency=2.4s2026.0368383.96510.5
Multi-Model PipelineLatency=2.8s2026.0362384.26112