Multimodal Search on MMSearch (Accuracy, Average Tool-call Turns)
88.5AccuracyHyperEyes-235B
Evaluation results
| Method | Links | ||
|---|---|---|---|
| HyperEyes-235BTraining Stage=RL, Evaluation Mode=Agentic Workflow2026.05 | 88.5 | 1.4 | |
| HyperEyes-30BTraining Stage=RL, Evaluation Mode=Agentic Workflow2026.05 | 86.9 | 1.6 | |
| Gemini-3.1-ProEvaluation Mode=Agentic Workflow2026.05 | 86.1 | 1.2 | |
| HyperEyes-235BTraining Stage=SFT, Evaluation Mode=Agentic Workflow2026.05 | 84.4 | 1.7 | |
| HyperEyes-30BTraining Stage=SFT, Evaluation Mode=Agentic Workflow2026.05 | 82 | 1.8 | |
| Kimi-K2.5Evaluation Mode=Agentic Workflow2026.05 | 76.6 | 2.2 | |
| Claude-Opus-4.6Evaluation Mode=Agentic Workflow2026.05 | 76.2 | 1.6 | |
| Gemini-3.1-ProEvaluation Mode=Direct Answer2026.05 | 75.4 | — | |
| REDSearchEvaluation Mode=Multimodal Deep Search Agents2026.05 | 72.9 | — | |
| VDREvaluation Mode=Multimodal Deep Search Agents2026.05 | 69.6 | 11.1 | |
| Kimi-K2.5Evaluation Mode=Direct Answer2026.05 | 65.6 | — | |
| Qwen3-VL-235BEvaluation Mode=Agentic Workflow2026.05 | 64.8 | 1.4 | |
| DeepEyes-V2Evaluation Mode=Multimodal Deep Search Agents2026.05 | 63.7 | 2.1 | |
| Claude-Opus-4.6Evaluation Mode=Direct Answer2026.05 | 59.8 | — | |
| WebWatcherEvaluation Mode=Multimodal Deep Search Agents2026.05 | 55.3 | 4.8 | |
| Qwen3-VL-30BEvaluation Mode=Agentic Workflow2026.05 | 54.1 | 1.7 | |
| MMSearch-R1Evaluation Mode=Multimodal Deep Search Agents2026.05 | 53.8 | 1.4 | |
| Qwen3-VL-235BEvaluation Mode=Direct Answer2026.05 | 30.3 | — | |
| Qwen3-VL-30BEvaluation Mode=Direct Answer2026.05 | 21.3 | — |