DeepSeek: R1 0528
May 28th update to the [original DeepSeek R1](/deepseek/deepseek-r1) Performance on par with [OpenAI o1](/openai/o1), but open-sourced and with fully open reasoning tokens. It's 671B parameters in size, with 37B active...
Everything else
| Max output | 33K |
|---|---|
| Input / 1M | $0.5 |
| Output / 1M | $2.15 |
| Cache read / 1M | $0.35 |
| Cache write / 1M | — |
| 1M in + 1M out | $2.65 |
| Takes | text |
| Returns | text |
| Tools | yes |
| Reasoning | always on |
| Open weights | yes |
| Released | 2025-05-28 |
| Knowledge cutoff | 2025-03-31 |
| Retires | — |
| Catalog id | deepseek/deepseek-r1-0528 |
Benchmarks
| source | epoch |
|---|---|
| gpqa | 0.763 |
| gpqaStderr | 0.024 |
| mathLevel5 | 0.966 |
| mathLevel5Stderr | 0.0042 |
| aime | 0.664 |
| aimeStderr | 0.0491 |
| simpleQA | 0.274 |
| simpleQAStderr | 0.0141 |
Benchmark scores by Epoch AI, CC BY 4.0.
Using it
In 00, this model is picked per agent — and per tier, so the model that answers your customers need not be the one that writes your code. Get 00.