DeepSeek's merged model combining deep reasoning with general instruction following at extremely low cost.
| Benchmark | Description | Score | Category | Bar |
|---|---|---|---|---|
| MMLU | Massive Multitask Language Understanding | 86.8% | knowledge | |
| HumanEval | Code generation accuracy | 89.2% | coding | |
| MATH | Mathematical reasoning | 86% | math | |
| GSM8K | Grade school math | 88% | math | |
| GPQA | Graduate-level science Q&A | 49.7% | reasoning | |
| HellaSwag | Common sense reasoning | 85.3% | reasoning | |
| SWE-bench | Software engineering problems | 36.4% | coding | |
| ARC-C | AI2 Reasoning Challenge | 93% | reasoning |