Multimodal long-horizon complex reasoning on RealXBench
35.6ScoreKimi-K2.5
Evaluation results
| Method | Links | |
|---|---|---|
| Kimi-K2.5Size=1T, Train=-2026.06 | 35.6 | |
| Gemini-3.1-Pro-PreviewSize=-, Train=-2026.06 | 34.5 | |
| GLM-5Size=744B, Train=-2026.06 | 34.5 | |
| Qwen3-235BSize=235B, Train=-2026.06 | 32.6 | |
| Doubao-2.0-ProSize=-, Train=-2026.06 | 32.5 | |
| Qwen3.5-397BSize=397B, Train=-2026.06 | 32 | |
| MiniMax-M2.7Size=230B, Train=-2026.06 | 32 | |
| S1-DeepResearchSize=32B, Train=SFT2026.06 | 31.4 | |
| Claude-4.6-Sonnet-ThinkingSize=-, Train=-2026.06 | 30.4 | |
| GPT-5.2Size=-, Train=-2026.06 | 30.4 | |
| Qwen3-32BSize=32B, Train=-2026.06 | 28.4 | |
| DeepSeek-V3.2Size=671B, Train=-2026.06 | 23.7 |