Reasoning on HLE (test)
26AccuracyMTRouter
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MTRouterRouting Level=Multi-Turn Routers (turn-level)2026.04 | 26 | 35 | |
| GPT-5Routing Level=Single-Model Baselines2026.04 | 25.1 | 61.8 | |
| EmbedLLMRouting Level=Single-Turn Routers (episode-level)2026.04 | 24.8 | 61.4 | |
| Router-R1Routing Level=Multi-Turn Routers (turn-level)2026.04 | 24.2 | 51.9 | |
| LLM RouterRouting Level=Multi-Turn Routers (turn-level)2026.04 | 24 | 56.2 | |
| AvengersProRouting Level=Single-Turn Routers (episode-level)2026.04 | 23.7 | 47.5 | |
| Random RouterRouting Level=Multi-Turn Routers (turn-level)2026.04 | 20 | 16.9 | |
| OpenRouterRouting Level=Multi-Turn Routers (turn-level)2026.04 | 18.3 | 138.5 | |
| DeepSeek-V3.2Routing Level=Single-Model Baselines2026.04 | 15.6 | 22.4 | |
| RouterDCRouting Level=Single-Turn Routers (episode-level)2026.04 | 12.8 | 10.3 | |
| Kimi-K2Routing Level=Single-Model Baselines2026.04 | 11.4 | 12 | |
| GPT-OSS-120BRouting Level=Single-Model Baselines2026.04 | 9.7 | 0.7 | |
| MiniMax-M2Routing Level=Single-Model Baselines2026.04 | 7.8 | 18.1 | |
| Gemini-2.5-Flash-LiteRouting Level=Single-Model Baselines2026.04 | 5.6 | 3 |