Temporal Regret Control on CAP-GSM8K GPT Claude Frontier Models (E=100)
2.3RCE_tempTrivium
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| TriviumModel=Claude-Sonnet-4.5, Horizon (E)=100, seeds=3, API=Anthropic API2026.06 | 2.3 | 0.5 | 100 | 8 | |
| TriviumModel=GPT-4o, Horizon (E)=100, seeds=3, API=OpenAI API2026.06 | 3 | 0.65 | 100 | 6 | |
| TriviumModel=GPT-3.5-Turbo, Horizon (E)=100, seeds=3, API=OpenAI API2026.06 | 3.7 | 0.8 | 100 | 5 | |
| ReactiveModel=Claude-Sonnet-4.5, Horizon (E)=100, seeds=3, API=Anthropic API2026.06 | 20 | 4.3 | 0 | — | |
| RLVRModel=Claude-Sonnet-4.5, Horizon (E)=100, seeds=3, API=Anthropic API2026.06 | 21.3 | 4.6 | 0 | — | |
| ReactiveModel=GPT-4o, Horizon (E)=100, seeds=3, API=OpenAI API2026.06 | 57.3 | 12.4 | 0 | — | |
| RLVRModel=GPT-4o, Horizon (E)=100, seeds=3, API=OpenAI API2026.06 | 58 | 12.6 | 0 | — | |
| RLVRModel=GPT-3.5-Turbo, Horizon (E)=100, seeds=3, API=OpenAI API2026.06 | 68 | 14.8 | 0 | — | |
| ReactiveModel=GPT-3.5-Turbo, Horizon (E)=100, seeds=3, API=OpenAI API2026.06 | 68.7 | 14.9 | 0 | — |