DeepSeek R1 is the model that made open-weight reasoning mainstream. The January 2025 release — and the stronger R1-0528 update — showed that RL-trained chain-of-thought reasoning could be published under MIT and still compete with proprietary reasoning models on math and code.
By mid-2026 R1 is no longer DeepSeek's frontier, but it remains historically important and practically useful: its distilled variants (1.5B to 70B) are still among the most-downloaded reasoning models on Hugging Face, and 'run DeepSeek R1 locally' remains the entry point through which many developers first self-host a serious model.
DeepSeek R1 specifications
| Vendor | DeepSeek (China) |
|---|---|
| Architecture | Mixture-of-Experts reasoning model (RL-trained chain-of-thought) |
| Total parameters | 671B |
| Active parameters | 37B |
| Context window | 128K tokens |
| Modalities | text |
| License | MIT — weights free to download, self-host, fine-tune, and use commercially |
| Release date | 2025-05-28 |
| Weights | Hugging Face · GitHub |
Benchmarks and reported results
| Benchmark | Result | Note |
|---|---|---|
| AIME 2025 | 87.5% | R1-0528, with thinking |
Running DeepSeek R1 locally
The full 671B model needs a multi-GPU server (~350–400GB at 4-bit). The official distilled variants (1.5B–70B, including Qwen and Llama bases) are the local-deployment path — the 32B distill runs on a single 24GB GPU quantized.
What DeepSeek R1 is best for
- Open-weight math and logic reasoning with visible chains of thought
- Local deployment via the distilled 7B–70B variants
- Research on RL-trained reasoning (fully published methodology)
Frequently asked questions
What hardware do I need to run DeepSeek R1 locally?
The full 671B model needs roughly 350–400GB of memory at 4-bit quantization — server territory. The official distills are the local path: R1-Distill-Qwen-32B runs on a single 24GB GPU (RTX 3090/4090) at 4-bit, and the 7B/8B distills run on 8GB GPUs or Apple Silicon laptops.
Has DeepSeek R2 been released?
No. As of July 2026 there is no official R2 announcement, no R2 API entry, and no model card. DeepSeek's 2026 flagship turned out to be the V4 family (April 2026), which integrates reasoning rather than shipping a separate R-series release.