Public preview · figures are illustrative fixture data, attributed to their source · data to May 2025 · read-only
Model · DeepSeek-R1 · January 2025 Live · Hugging Face

DeepSeek-R1 671B

A 671 billion-parameterHow many numbers the model learned during training — the usual rough measure of its size. mixture-of-expertsA model split into parts where only a few run for each word, so it answers faster than its size suggests. model using ~37 billion parameters per token from DeepSeek. Open reasoning model plus distilled dense models.

Size
671B37B active
Memory to run
~169 GBsmallest, 8K ctx
Context
160Ktokens
License
PermissiveMIT License
Updated
18 Sept 2026

Which version to use

  • Variant DeepSeek-R1 · DeepSeek
    Trained to think step by step before answering; slower, stronger on hard problems.
    ChatReasoningCodingLong context

Where it runs

Best-fitting version on each reference system, at an 8K contextHow much text the model can consider at once, counted in tokens — roughly ¾ of a word each..

Benchmarks

Each benchmarkA fixed set of questions every model is given, so their scores can be compared on the same task. is developer-reported; prompts and settings differ between labs. Bars are relative to the best open result.

BenchmarkDeepSeek-R1
Benchmark GPQA DiamondAccuracy
71.5src
Benchmark LiveCodeBenchpass@1
65.9src
Benchmark MATH-500Accuracy
97.3src
Benchmark MMLU-ProAccuracy
84.0src

developer reported

Variants & downloads

Each variant is a separate set of weights. Expand one for its downloads: quantizationStoring each of the model’s numbers with fewer bits, so the file is smaller and needs less memory. trades a little quality for a much smaller file, and the memory column adds the working memoryExtra space the model needs while it answers. It grows with the length of the conversation, on top of the file itself. a conversation needs on top.

Variant DeepSeek-R1reasoning · DeepSeek · 3 downloads · MIT LicensePermissive
What it is
Trained to think step by step before answering; slower, stronger on hard problems.
Publisher
DeepSeek
License
MIT License · commercial use allowed
Released
20 Jan 2025

Contributions are not open yet, so there is nothing here from members.

QuantizationFormatBitsDownloadMemory @ 8KPublisher
Download FP8fp8safetensors8.1633 GiB~659.1 GiBDeepSeekdeepseek-ai/DeepSeek-R1
Download Q3_K_Mk quantgguf3.91305 GiB~318.7 GiBUnsloth AIunsloth/DeepSeek-R1-GGUF
Download IQ2_XXSi quantgguf2.06161 GiB~168.4 GiBUnsloth AIunsloth/DeepSeek-R1-GGUF
Architecture detailsLayers, attention and KV cache geometry
Parameters
671B
Active / token
37B
Architecture
Mixture of experts
Layers
61
Attention heads
128
KV heads
128
Head dim
128
KV cache @ 8K (fp16)
0.54 GiB
Max context
163,840 tokens

Measured performance

Throughput on specific systems and runtimes, with the source of each measurement.

No performance measurements recorded yet.

Community results

Contributions are not open yet, so there is nothing here from members.

Reviews

Contributions are not open yet, so there is nothing here from members.

Lineage

How this model’s variants relate to each other and to other models.

VariantDeepSeek-R1 reasoning
VariantDeepSeek-R1-Distill-Qwen-7B distilled from DeepSeek-R1
VariantDeepSeek-R1-Distill-Qwen-32B distilled from DeepSeek-R1

Sources & history

Sources

  • Hugging Face Hub (live source, 1 record, 18 Sept 2026)
  • Mutinai illustrative fixtures (illustrative fixture, 1 record, 13 Sept 2026)

External identifiers

Field history

  • license = "mit" Hugging Face Hub (current)

Timeline

  1. DeepSeek-R1 and distilled models released under MIT