DeepSeek-R1 vs DeepSeek-V3
A direct, empirical evaluation of DeepSeek-R1 (DeepSeek-AI) against DeepSeek-V3 (DeepSeek-AI).
DeepSeek-R1
Developed by DeepSeek-AI · 671B parameters
Olympiad mathematics, logic verification, and autonomous thinking chains.
DeepSeek-V3
Developed by DeepSeek-AI · 671B parameters
Cost-effective high-throughput general reasoning and bilingual generation.
| Metric | DeepSeek-R1 | DeepSeek-V3 |
|---|---|---|
| SWE-bench Verified | 49.2% | 49.2% |
| MMLU-Pro | 84% | 75.9% |
| MATH-500 | 97.3% | 90.2% |
| Context Window | 128,000 tokens | 128,000 tokens |
| Input Pricing (per 1M) | $0.55 | $0.14 |
| Output Pricing (per 1M) | $2.19 | $0.28 |