High-level Multimodal Reasoning on HLE-500
29.5Text ScoreAutoAgent
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| AutoAgentBackbone=DeepSeek-R12026.03 | 29.5 | 23.9 | 28.2 | |
| ReActBackbone=DeepSeek-R12026.03 | 15.2 | 11.5 | 14.4 | |
| Direct AnswerBackbone=DeepSeek-R12026.03 | 14.7 | 2.7 | 12 | |
| DeepAgentBackbone=DeepSeek-R12026.03 | 8.3 | 8 | 8.2 | |
| DeepAgentBackbone=Qwen3-30B-A3B2026.03 | 8 | 4.4 | 7.2 | |
| AutoAgentBackbone=Qwen3-30B-A3B2026.03 | 5.7 | 8 | 6.2 | |
| DeepAgentBackbone=QwQ-32B2026.03 | 4.9 | 2.7 | 4.4 | |
| Direct AnswerBackbone=Qwen3-30B-A3B2026.03 | 4.9 | 4.4 | 4.8 | |
| AutoAgentBackbone=QwQ-32B2026.03 | 4.7 | 5.3 | 4.8 | |
| ReActBackbone=QwQ-32B2026.03 | 4.4 | 1.8 | 3.8 | |
| Direct AnswerBackbone=QwQ-32B2026.03 | 4.1 | 2.7 | 3.8 | |
| ReActBackbone=Qwen3-30B-A3B2026.03 | 4.1 | 4.4 | 4.2 |