The Open-Source AI Stack
RSS
All models

Models · deepseek

DeepSeek-V3.2

Open DeepSeek · — · MIT

Successor to V3.2-Exp, released December 1 2025 with DeepSeek Sparse Attention for long-context efficiency. Integrated reasoning with tool use through a new chat template, and a Speciale variant claimed gold-level finishes on IMO, CMO, ICPC, and IOI 2025.

Cost

$0.50 / Mtok input
$1.60 / 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 Attention (not disclosed) RoPE + YaRN context not disclosed 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
not disclosed
Active params
37B
Context window
not verified
Attention
dsa
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 83.7 as of 2026-05-21 source ↗
GPQA-Diamond 75.1 as of 2026-05-21 source ↗

Code

LiveCodeBench 59.3 as of 2026-05-21 source ↗

Math

AIME 2025 59.0 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
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

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

  • · DeepSeek Sparse Attention (DSA)
  • · Thinking with tools via new chat template
  • · V3.2-Speciale gold-level math and coding olympiad results

Lineage

Successor to the V3.2-Exp checkpoint (not catalogued; production V3.2 released 2025-12-01). Introduces DeepSeek Sparse Attention; Speciale variant aimed at math olympiads.

Derivatives

Sources