Competition Math on IMOBench
52.1AccuracySELFCOMPACT
Evaluation results
| Method | Links | |
|---|---|---|
| SELFCOMPACTModel=Qwen3-30B-A3B-Instruct-2507, Avg. tokens per question=29k, Thinking state=Enabled2026.06 | 52.1 | |
| Fixed Interval SummaryModel=Qwen3-30B-A3B-Instruct-2507, Avg. tokens per question=26k, Thinking state=Enabled2026.06 | 48.7 | |
| SELFCOMPACTModel=Qwen3-4B-Instruct-2507, Avg. tokens per question=48k, Thinking state=Enabled2026.06 | 45.5 | |
| No CompactionModel=Qwen3-30B-A3B-Instruct-2507, Avg. tokens per question=16k, Thinking state=Enabled2026.06 | 45.2 | |
| Fixed Interval SummaryModel=Qwen3-4B-Instruct-2507, Avg. tokens per question=44k, Thinking state=Enabled2026.06 | 41.4 | |
| SELFCOMPACTModel=Qwen3.5-9B, Avg. tokens per question=93k, Thinking state=Disabled2026.06 | 41.4 | |
| No CompactionModel=Qwen3-4B-Instruct-2507, Avg. tokens per question=16k, Thinking state=Enabled2026.06 | 38.9 | |
| SELFCOMPACTModel=Qwen3.5-4B, Avg. tokens per question=67k, Thinking state=Disabled2026.06 | 34.1 | |
| Fixed Interval SummaryModel=Qwen3.5-9B, Avg. tokens per question=90k, Thinking state=Disabled2026.06 | 33.4 | |
| Fixed Interval SummaryModel=Qwen3.5-4B, Avg. tokens per question=64k, Thinking state=Disabled2026.06 | 27.4 | |
| No CompactionModel=Qwen3.5-9B, Avg. tokens per question=16k, Thinking state=Disabled2026.06 | 25 | |
| No CompactionModel=Qwen3.5-4B, Avg. tokens per question=16k, Thinking state=Disabled2026.06 | 16.9 |