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

Downstream Accuracy on LiveCodeBench

33.2Accuracy

Victim-Trace (oracle)

Updated 1mo ago

Evaluation Results

Method	Links
Victim-Trace (oracle) 2026.03		33.2
Synthesized-Trace 2026.03		30.9
No finetuning 2026.03		28.9
Answer+Summary 2026.03		25.6
Answer-only 2026.03		25.4
Surrogate-Trace 2026.03		22.9
No finetuning 2026.03		13.1
Answer+Summary 2026.03		11.7
Victim-Trace (oracle) 2026.03		10.4
Answer-only 2026.03		9.6
Surrogate-Trace 2026.03		9.1
Synthesized-Trace 2026.03		8.8