DeepSeek R1 671B
By DeepSeek · China
Updated 2026-07-13
Overview
The reference open reasoning model — a 671B MoE with 37B active, released under MIT. Scores 97.3 on MATH-500, 79.8 on AIME, and 90.8 on MMLU.
When to pick this model
- You're running a dedicated inference server and need frontier reasoning
- You want the strongest open math, code, and logic model available
- You need an MIT-licensed model with no commercial restrictions
- You're benchmarking against closed frontier models like o1 or o3
VRAM requirements by quantization
| Quantization | VRAM required |
|---|---|
| Q4_K_M (recommended) | 400 GB |
| Q5_K_M | 480 GB |
| Q8_0 | 720 GB |
| FP16 (no quantization) | 1342 GB |
VRAM figures include model weights plus a typical 8k KV cache and ~600 MB runtime overhead (Ollama / llama.cpp baseline). Add headroom for higher context lengths.
In practice, DeepSeek R1 671B is server-class even at Q4_K_M (400 GB). Stepping up to Q8_0 nearly doubles the footprint to 720 GB, and unquantized FP16 weights take 1342 GB — plan your GPU around the Q4 or Q5 figure unless you specifically need the higher fidelity.
Without a GPU, DeepSeek R1 671B needs roughly 512 GB of system RAM to run on CPU via llama.cpp or Ollama — workable for background jobs, but far slower than GPU inference. Throughput estimates from our compatibility engine: around 1 tokens/sec on entry-level GPUs, on the order of 5 tokens/sec on a mid-range card, and up to 15 tokens/sec on high-end hardware — assuming the chosen quantization fully fits in VRAM.
What hardware do you need
The table below matches DeepSeek R1 671B to common GPU memory tiers, using the highest-fidelity quantization that fully fits each card class. Spilling layers to system RAM works but costs most of the speed, so size your card to the quantization you actually want to run.
| GPU memory | Example cards | Best fit for DeepSeek R1 671B |
|---|---|---|
| 8 GB | RTX 5070 Laptop, RTX 5060, RTX 5060 Ti 8GB | Does not fit — needs 400 GB at Q4_K_M |
| 12 GB | RTX 5070, RTX 5070 Ti Laptop, RTX 4080 Laptop | Does not fit — needs 400 GB at Q4_K_M |
| 16 GB | RTX 5080, RTX 4080 Super, Radeon RX 9070 XT | Does not fit — needs 400 GB at Q4_K_M |
| 24 GB | RTX 4090, Radeon RX 7900 XTX, RTX 5090 Laptop | Does not fit — needs 400 GB at Q4_K_M |
| 32 GB | RTX 5090 | Does not fit — needs 400 GB at Q4_K_M |
Which GPU should you buy to run DeepSeek R1 671B?
To run DeepSeek R1 671B locally at Q4, you need ~400 GB of VRAM. The best value for this is a Apple Mac Studio (64+ GB unified memory).
As an Amazon Associate, BestLLMfor earns from qualifying purchases, at no extra cost to you. It does not influence our independent rankings.
Published benchmark scores
| Benchmark | Score |
|---|---|
| MMLU | 90.8 |
| GPQA Diamond | 71.5 |
| MATH-500 | 97.3 |
Scores published by the model author or aggregated from public leaderboards. Re-measured monthly by our editorial team.
To put DeepSeek R1 671B in context: its MMLU score of 90.8 ranks #1 of the 34 catalog models with a published MMLU result (catalog median 73.4); its MATH-500 score of 97.3 ranks #1 of the 8 catalog models with a published MATH-500 result (catalog median 93.3). Scores are author-published and measured under different harnesses, so treat gaps of a few points as noise rather than a verdict.
Strengths
- MIT license — no commercial restrictions
- Reference open reasoning model
- MATH-500 score of 97.3
- R1-0528 update further sharpens reasoning
Limitations
- 400GB+ in Q4 — server-class hardware required
- Out of reach for any single-machine local setup
- Very long reasoning traces drive up latency
Typical workloads
In our catalog grid, DeepSeek R1 671B is filed under Frontier Reasoning, Math/Science — the use cases where its size/quality trade-off makes the most sense. Its tags translate to concrete workloads: multi-step reasoning and math-flavoured tasks.
The 125k-token context window covers long chats and mid-sized documents, though very large retrieval workloads will need chunking. The MIT license is permissive, so shipping it inside a commercial product raises no special legal questions.
Architecture & training
Architecture: MoE (inherited from V3) · Multi-head Latent Attention · auxiliary-loss-free · RL-trained
Training: Distillation + multi-stage RL. R1-0528 update (May 2025).
The open reasoning gold standard — if you have the hardware to host it.
Quick start
ollama run deepseek-r1:671bOr use the open-source MCP server to query this model from Claude Desktop, Cursor, or any MCP-compatible client.
Similar models worth comparing
Frequently asked questions
How much VRAM does DeepSeek R1 671B need?
At the recommended Q4_K_M quantization, DeepSeek R1 671B needs about 400 GB of VRAM. Q8_0 takes 720 GB, and unquantized FP16 weights take 1342 GB.
Can DeepSeek R1 671B run without a GPU?
Yes — with roughly 512 GB of system RAM it runs CPU-only through llama.cpp or Ollama. Expect a fraction of GPU speed, which is fine for background or batch jobs but slow for interactive chat.
What context window does DeepSeek R1 671B support?
DeepSeek R1 671B supports a 125k-token context window (128,000 tokens).
Can I use DeepSeek R1 671B commercially?
Yes. DeepSeek R1 671B is released under MIT, a permissive open-source license that allows commercial use, modification and redistribution.
How fast is DeepSeek R1 671B on consumer hardware?
Our compatibility engine estimates on the order of 5 tokens/sec on a mid-range GPU and up to 15 tokens/sec on high-end cards, assuming the quantization fully fits in VRAM.
Which quantization of DeepSeek R1 671B should I download first?
Start with Q4_K_M (400 GB) — the standard size/quality sweet spot. Step up to Q5_K_M or Q8_0 only if you have VRAM headroom. It does not fit a single 24 GB consumer card — plan for multi-GPU or server hardware.