List your company

Long-context QA on RULER 1200 (test)

84.1Success RateCVT-RL
76.09278.17180.2582.329Jun 3, 2026

Evaluation results

MethodLinks
CVT-RLDescription=per-seed CVT-RL success, Split (train/dev/test)=7200/900/1200, Individual Seeds=83.6/84.7/82.9/84.4/84.82026.0684.14.7
IMDescription=information-matched counterfactual-process RL, Split (train/dev/test)=7200/900/12002026.0679.4—
CMDescription=compute-matched non-causal RL, Split (train/dev/test)=7200/900/12002026.0676.4—