ResearchBenchmarksTemporal Regret Control on CAP-GSM8K E=500 Llama-3.3-70bTemporal Regret Control on CAP-GSM8K E=500 Llama-3.3-70bFollow2.4RCE_temp(E)Trivium-7.62460.038127.7195.362Jun 3, 2026Evaluation resultsMethodMethodLinksRCE_temp(E)Alpha Hat (α^)Commit RateVerdict MarginTriviumModel=Llama-3.3-70b, H...Model=Llama-3.3-70b, Horizon (E)=500, seeds=10, API=Groq API2026.062.40.3910010RLVRModel=Llama-3.3-70b, H...Model=Llama-3.3-70b, Horizon (E)=500, seeds=10, API=Groq API2026.06248.8400—ReactiveModel=Llama-3.3-70b, H...Model=Llama-3.3-70b, Horizon (E)=500, seeds=10, API=Groq API2026.0625340.70—