Share your thoughts, 1 month free Claude Pro on usSee more

Agentic Tool-use on AppWorld Normal

89.3Task Goal Completion (TGC)

RCL (all primitives)

Updated 13d ago

Evaluation Results

Method	Links
RCL (all primitives) 2026.04		89.3
ACE + Auxiliary Losses 2026.04		89.2
RCL (all primitives) 2026.04		89.1
ACE + Batching 2026.04		88.7
ACE + Optimizer State 2026.04		88.1
ACE + Batching 2026.04		87.2
ACE 2026.04		86.9
ACE + Grouped Rollouts 2026.04		86.8
ACE + Grouped Rollouts 2026.04		86.3
ACE + Auxiliary Losses 2026.04		86.3
ACE + Failure Replay 2026.04		84.5
ACE + Optimizer State 2026.04		84.5
ACE + Credit Assignment 2026.04		84.5
ACE + Credit Assignment 2026.04		83.9
ACE 2026.04		83.3
GEPA 2026.04		82.7
ACE + Failure Replay 2026.04		82.7
Seed 2026.04		81.5
Seed 2026.04		78
GEPA 2026.04		76.2