The Open-Source AI Stack
RSS
All models

Models · deepseek

DeepSeek-V3.1

Open DeepSeek · — · MIT

Hybrid V3.1 with one model serving both thinking and non-thinking modes, released August 21 2025. DeepSeek added 840B tokens of continued pretraining for long-context extension (32K and 128K phases) and shipped under MIT. Pricing update took effect September 5 2025.

Cost

$0.56 / Mtok input
$1.67 / Mtok output

· as of 2026-05-21

source ↗

Speed

0 tok/sec output
0 ms TTFT

· as of 2026-05-21

source ↗

Architecture

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

Specs

Architecture
moe
Total params
671B
Active params
37B
Context window
128K tokens
Attention
mla
Position encoding
rope-yarn
Post-training
sft, rlhf
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 83.3 as of 2026-05-21 source ↗
GPQA-Diamond 73.5 as of 2026-05-21 source ↗

Code

LiveCodeBench 57.7 as of 2026-05-21 source ↗

Math

AIME 2025 49.7 as of 2026-05-21 source ↗

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

  • · Single checkpoint with hybrid thinking and non-thinking modes
  • · UE8M0 FP8 scale format for microscaling-compatible weights
  • · 840B-token long-context extension on top of V3 base

Lineage

Hybrid V3.1 merges DeepSeek's chat and reasoning tracks behind one model.

Derived from

DeepSeek-V3 2024-12-26

Derivatives

Sources