DeepSeek-R1 is a reasoning-focused model that delivers exceptional performance relative to its cost. At just $0.55/$2.19 per MTok, it scores 1418 on the LMSYS Arena, outperforming many models that cost 10x more. The R1 series introduced the thinking/reasoning approach that was later adopted by other labs. It is available as open weights for self-hosting.
| Benchmark | Score |
|---|---|
| LMSYS Arena | 1418 |
| Context Window | 128K |
| Price (input/output) | $0.55/$2.19 |