The Open-Source AI Stack
RSS
All models

Models · deepseek

DeepSeek-R1 (May 2025 refresh)

Open DeepSeek · 2025-05-28 · MIT

An RL-only refresh of R1 that gained substantial ground on reasoning benchmarks (notably AIME 2024) without any new pretraining. Tightened the open-vs-closed reasoning gap.

Cost

— / Mtok input
— / Mtok output

DeepSeek API · as of 2026-05-19

via Artificial Analysis ↗

Speed

— tok/sec output

DeepSeek API · as of 2026-05-19

via Artificial Analysis ↗

Architecture

tokens in Embedding vocab not disclosed · deepseek tokenizer × N layers Multi-head Latent Attention RoPE + YaRN context 128,000 tokens MoE Router 256 experts total · 8 active per token shown: 32 of 256 Output projection tokens out
Schema-generated from data/models.yaml. Every label is auditable against the model's sources.

Specs

Architecture
moe
Total params
not disclosed
Active params
37B
Experts
256 total · 8 active
Context window
not verified
Attention
mla
Position encoding
rope-yarn
Training hardware
H800
Post-training
sft, grpo, rejection-sampling
OSI-approved
yes
Data released
no
Training code
not released

Benchmarks

Each score carries the date it was published; we never infer or interpolate missing scores.

General reasoning

MMLU-Pro 85.0 as of 2025-05-28 source ↗
GPQA-Diamond 81.0 as of 2025-05-28 source ↗

Code

LiveCodeBench 73.3 as of 2025-05-28 source ↗

Math

AIME 2024 91.4 as of 2025-05-28 source ↗

Recommended use cases

  • math reasoning
  • code reasoning
  • agentic workflows

Available quantizations

GGUF llama.cpp's container; the common local format, k-quants from Q2 to Q8. runs on llama.cpp, Ollama
AWQ Activation-aware 4-bit weight quantization for GPU serving. runs on vLLM, SGLang
GPTQ Post-training 4-bit weight quantization for GPU serving. runs on vLLM, SGLang, Transformers
MLX Apple MLX 4/8-bit layout for Apple silicon. runs on Apple MLX
FP8 8-bit float, frequently a native release on Hopper / Blackwell GPUs. runs on vLLM, SGLang, TensorRT-LLM
bitsandbytes On-the-fly NF4 / INT8 weight quantization inside Transformers. runs on Transformers

Verified via the Hugging Face model tree ↗. Community quantizations change over time; the families shown are those with published weights at audit time.

Notable innovations

  • · RL-only update path
  • · Notable reasoning gains without new pretrain

Lineage

RL-only refresh; no new pretrain.

Derived from

DeepSeek-R1 2025-01-20

Sources