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

Agentic Tool-use on AppWorld (Challenge)

83.7TGC

RCL (all primitives)

Updated 13d ago

Evaluation Results

Method	Links
RCL (all primitives) 2026.04		83.7
ACE + Batching 2026.04		83.4
ACE + Failure Replay 2026.04		81.8
ACE + Grouped Rollouts 2026.04		81.4
ACE 2026.04		80.7
ACE + Auxiliary Losses 2026.04		80.6
ACE + Optimizer State 2026.04		79.4
ACE + Credit Assignment 2026.04		79.4
Seed 2026.04		75.8
ACE + Auxiliary Losses 2026.04		74.6
ACE + Optimizer State 2026.04		73.9
ACE + Failure Replay 2026.04		73.1
ACE + Grouped Rollouts 2026.04		72.4
RCL (all primitives) 2026.04		71.9
ACE + Credit Assignment 2026.04		71.5
ACE + Batching 2026.04		71.2
ACE 2026.04		69.1
GEPA 2026.04		66.7
Seed 2026.04		64.3
GEPA 2026.04		62.6