A fast, experimentally-minded model that handles both text and images within an enormous one-million-token context window. It sits in DeepSeek's lineup as a lighter, quicker option — trading some depth for speed and efficiency. The 'Exp' tag signals it's still evolving, so expect rough edges alongside its broad multimodal reach.
| Benchmark | Score | Type | Recorded |
|---|---|---|---|
| MMLU-Pro | 86.2 | accuracy | 29d ago |
| SWE-Bench | 79.0 | accuracy | 29d ago |
| GPQA Diamond | 88.1 | accuracy | 29d ago |
| LiveCodeBench | 87.3 | accuracy | 29d ago |
| LCR | 79.7 | accuracy | 29d ago |
| Humanity's Last Exam | 34.8 | accuracy | 29d ago |