Open-source LLM · DeepSeek
DeepSeek R1
671 B parameters (37 B active per token) · 128K context · MIT · released 2025-01-20 by DeepSeek.
DeepSeek R1 needs about 1342 GB at FP16: no single GPU can hold it. Serving requires an 8-GPU HGX baseboard (INT8/INT4) or a full rack (NVL72, TPU pod) at FP16.
Where this model sits
The reasoning variant of V3. Long chain-of-thought traces mean actual token cost per query is 5-20x V3's, but quality on math + code jumps into o1 territory.
What runs it (production serving)
Datacenter GPUs and rack designs sized for DeepSeek R1. Consumer cards are not practical at this size.
Smallest datacenter GPU that fits (INT4)
Multi-GPU / rack-scale (FP16)
Memory at different context lengths
Weights are fixed. The KV cache grows linearly with context length and can rival the weights at full window. Most guides only quote weights, then hand-wave a 15% overhead, which is only true at short context.
| Context | FP16 total | INT8 total | INT4 total |
|---|---|---|---|
| 4K (short chat) | 1342 + 15.3 = 1357.3 GB | 671 + 7.6 = 678.6 GB | 336 + 7.6 = 343.1 GB |
| 32K (typical RAG) | 1342 + 122.0 = 1464.0 GB | 671 + 61.0 = 732.0 GB | 336 + 61.0 = 396.5 GB |
| 128K (full window) | 1342 + 488.0 = 1830.0 GB | 671 + 244.0 = 915.0 GB | 336 + 244.0 = 579.5 GB |
KV bytes per token = 2 × 61 layers × 128 KV heads × 128 head-dim × precision bytes. INT4 rows keep the KV cache at INT8 (typical serving default).
Quality benchmarks
Vendor-reported scores from the release blog or tech report. Every row cites its source.
| Benchmark | Score | Source | Date |
|---|---|---|---|
| MMLU (5-shot) | 90.8 | DeepSeek R1 tech report | 2025-01-20 |
| MATH (AIME 2024, pass@1) | 79.8 | DeepSeek R1 tech report | 2025-01-20 |
| GPQA (Diamond) | 71.5 | DeepSeek R1 tech report | 2025-01-20 |
| Codeforces (Elo) | 2029.0 | DeepSeek R1 tech report | 2025-01-20 |
Speed and cost to serve
What cloud providers currently charge to serve DeepSeek R1. Tokens per second is what the provider publishes on their own page. Batch size, prompt length, and inference engine can shift these numbers a lot; treat as directional.
| Provider | Hardware | Speed | $/M in | $/M out | As of |
|---|---|---|---|---|---|
| DeepSeek API | H800 | — | $0.55 | $2.19 | 2025-08-01 |
| Together AI | H100 nodes | — | $3.00 | $7.00 | 2025-08-01 |
| Fireworks AI | H100 nodes | — | $3.00 | $8.00 | 2025-08-01 |
Cheapest option in this list for output-heavy workloads: DeepSeek API at $2.19/M out. Fastest published: not enough data.
License and variants
- License
- MIT
- The gotcha
- Same MIT + DeepSeek use-case restrictions as V3.
- Variants
- Reasoning · Distills (Llama, Qwen)
- Context window
- 131,072 tokens (128K)
- Architecture
- MoE (671B total, 37B active per token)
Compare with
Model card: https://huggingface.co/deepseek-ai/DeepSeek-R1 · Release: DeepSeek announcement
See also: every open-source LLM we track · all AI chips · rack designs.