Skip to content
Weights Not Walls
Menu

deepseek-ai

DeepSeek-V3.2

Reviewed
Size
685B parameterscounted from the safetensors header on the Hub
Licence
unknown, MIT LicenseOur reading, check the licence: You can use, change and redistribute the model for any purpose, including commercially. Keep the copyright notice and the licence text with copies. You do not have to share your changes.
On the Hub since
1 Dec 2025the date the repo was created
Context length
160k tokensfrom the model's config
Fine-tune
of deepseek-ai/DeepSeek-V3.2-Exp-BaseDeepseekV32ForCausalLM
On Hugging Face
deepseek-ai/DeepSeek-V3.210.1M downloads in the last 30 days, 1.6k likes

Scores

11 independent results on 11 eval sets
  • default setting22.1%±4.122 Dec 2025, Epoch AI

Epoch's name: fireworks/deepseek-v3p2. Best closed model on this set: gpt-5.5-pro-pre-release_high, 52.4% (OpenAI). All 101 results

  • default setting2.1%±2.116 Dec 2025, Epoch AI

Epoch's name: fireworks/deepseek-v3p2. Best closed model on this set: gdm-ai-co-mathematician, 47.9% (Google DeepMind). All 72 results

APEX-AgentsReported

Epoch's name: deepseek/deepseek-v3.2. Best closed model on this set: gemini-4-carbon_unknown, 84.2% (Google DeepMind). All 66 results

ARC-AGIReported

Epoch's name: deepseek/deepseek-v3.2. Best closed model on this set: claude-fable-5_max, 98.5% (Anthropic). All 318 results

ARC-AGI-2Reported

Epoch's name: deepseek/deepseek-v3.2. Best closed model on this set: gpt-6-astra_max, 95.0% (OpenAI). All 299 results

CL-benchReported

Epoch's name: deepseek/deepseek-v3.2. Best closed model on this set: gpt-5.4-2026-03-05_xhigh, 27.9% (OpenAI). All 23 results

Epoch's name: deepseek/deepseek-v3.2. Best closed model on this set: gpt-5.5_high, 22.2% (OpenAI). All 17 results

DTBenchReported

Epoch's name: deepseek/deepseek-v3.2. Best closed model on this set: claude-fable-5-2_unknown, 99.2% (Anthropic). All 241 results

LMCAReported

Epoch's name: deepseek/deepseek-v3.2. Best closed model on this set: claude-fable-5-2_xhigh, 70.6% (Anthropic). All 203 results

ProofBenchReported
  • Reasoning effort: none8.0%±2.7vals.ai

Epoch's name: fireworks/deepseek-v3p2. Best closed model on this set: claude-fable-5-1_max, 100.0% (Anthropic). All 85 results

  • Agent: Terminus 239.6%±2.81 Dec 2025, tbench.ai

Epoch's name: deepseek/deepseek-v3.2. Best closed model on this set: gpt-5.5_unknown, 84.7% (OpenAI). All 202 results

Claimed by the maker Claimed

The model card has no claimed scores in a machine-readable form (the card's model-index), so none are shown. The maker may give scores in the card text: read the card.

Other views, which we link to and do not copy: look this model up on Artificial Analysis and the Arena leaderboard.

Independent scores from Epoch AI (CC-BY 4.0), fetched 19 Nov 2026: Epoch AI, ‘Capabilities & benchmarking’. Published online at epoch.ai. Retrieved from ‘https://epoch.ai/benchmarks’ [online resource]. Each row links its own source. What the badges mean.

Run it yourself

file sizes read from the Hub 19 Nov 2026

Memory needed = weights + context cache + 0.50 GB for the runtime. Context cache at 8k tokens: 0.54 GB, worked out from the model's config (deepseek-ai/DeepSeek-V3.2), at 16-bit precision. Choose a machine above to see a verdict for each format. Method

GGUF for llama.cpp, Ollama, LM Studio and others

unsloth/DeepSeek-V3.2-GGUF (converted by unsloth), 17k downloads. Sizes are the real files on the Hub, all parts added up.

FormatSizeMemory neededGet it
UD-IQ3_XXS254.4 GBfile size, 6 parts255.4 GB254.4 + 0.54 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/DeepSeek-V3.2-GGUF:UD-IQ3_XXS
Download
hf download unsloth/DeepSeek-V3.2-GGUF --include "UD-IQ3_XXS/DeepSeek-V3.2-UD-IQ3_XXS-*-of-00006.gguf" --local-dir DeepSeek-V3.2-GGUF
Q3_K_S270.5 GBfile size, 6 parts271.5 GB270.5 + 0.54 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/DeepSeek-V3.2-GGUF:Q3_K_S
Download
hf download unsloth/DeepSeek-V3.2-GGUF --include "Q3_K_S/DeepSeek-V3.2-Q3_K_S-*-of-00006.gguf" --local-dir DeepSeek-V3.2-GGUF
Q3_K_M298.2 GBfile size, 7 parts299.3 GB298.2 + 0.54 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/DeepSeek-V3.2-GGUF:Q3_K_M
Download
hf download unsloth/DeepSeek-V3.2-GGUF --include "Q3_K_M/DeepSeek-V3.2-Q3_K_M-*-of-00007.gguf" --local-dir DeepSeek-V3.2-GGUF
UD-Q3_K_XL299.0 GBfile size, 7 parts300.0 GB299.0 + 0.54 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/DeepSeek-V3.2-GGUF:UD-Q3_K_XL
Download
hf download unsloth/DeepSeek-V3.2-GGUF --include "UD-Q3_K_XL/DeepSeek-V3.2-UD-Q3_K_XL-*-of-00007.gguf" --local-dir DeepSeek-V3.2-GGUF
IQ4_XS333.7 GBfile size, 8 parts334.8 GB333.7 + 0.54 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/DeepSeek-V3.2-GGUF:IQ4_XS
Download
hf download unsloth/DeepSeek-V3.2-GGUF --include "IQ4_XS/DeepSeek-V3.2-IQ4_XS-*-of-00008.gguf" --local-dir DeepSeek-V3.2-GGUF
IQ4_NL353.1 GBfile size, 8 parts354.1 GB353.1 + 0.54 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/DeepSeek-V3.2-GGUF:IQ4_NL
Download
hf download unsloth/DeepSeek-V3.2-GGUF --include "IQ4_NL/DeepSeek-V3.2-IQ4_NL-*-of-00008.gguf" --local-dir DeepSeek-V3.2-GGUF
Q4_0354.0 GBfile size, 8 parts355.0 GB354.0 + 0.54 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/DeepSeek-V3.2-GGUF:Q4_0
Download
hf download unsloth/DeepSeek-V3.2-GGUF --include "Q4_0/DeepSeek-V3.2-Q4_0-*-of-00008.gguf" --local-dir DeepSeek-V3.2-GGUF
Q4_K_S354.9 GBfile size, 8 parts355.9 GB354.9 + 0.54 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/DeepSeek-V3.2-GGUF:Q4_K_S
Download
hf download unsloth/DeepSeek-V3.2-GGUF --include "Q4_K_S/DeepSeek-V3.2-Q4_K_S-*-of-00008.gguf" --local-dir DeepSeek-V3.2-GGUF
Q4_K_M377.6 GBfile size, 9 parts378.6 GB377.6 + 0.54 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/DeepSeek-V3.2-GGUF:Q4_K_M
Download
hf download unsloth/DeepSeek-V3.2-GGUF --include "Q4_K_M/DeepSeek-V3.2-Q4_K_M-*-of-00009.gguf" --local-dir DeepSeek-V3.2-GGUF
UD-Q4_K_XL379.8 GBfile size, 9 parts380.8 GB379.8 + 0.54 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/DeepSeek-V3.2-GGUF:UD-Q4_K_XL
Download
hf download unsloth/DeepSeek-V3.2-GGUF --include "UD-Q4_K_XL/DeepSeek-V3.2-UD-Q4_K_XL-*-of-00009.gguf" --local-dir DeepSeek-V3.2-GGUF
Q4_1391.9 GBfile size, 9 parts392.9 GB391.9 + 0.54 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/DeepSeek-V3.2-GGUF:Q4_1
Download
hf download unsloth/DeepSeek-V3.2-GGUF --include "Q4_1/DeepSeek-V3.2-Q4_1-*-of-00009.gguf" --local-dir DeepSeek-V3.2-GGUF
Q5_K_S430.9 GBfile size, 10 parts431.9 GB430.9 + 0.54 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/DeepSeek-V3.2-GGUF:Q5_K_S
Download
hf download unsloth/DeepSeek-V3.2-GGUF --include "Q5_K_S/DeepSeek-V3.2-Q5_K_S-*-of-00010.gguf" --local-dir DeepSeek-V3.2-GGUF
Q5_K_M443.5 GBfile size, 10 parts444.5 GB443.5 + 0.54 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/DeepSeek-V3.2-GGUF:Q5_K_M
Download
hf download unsloth/DeepSeek-V3.2-GGUF --include "Q5_K_M/DeepSeek-V3.2-Q5_K_M-*-of-00010.gguf" --local-dir DeepSeek-V3.2-GGUF
UD-Q5_K_XL448.8 GBfile size, 10 parts449.8 GB448.8 + 0.54 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/DeepSeek-V3.2-GGUF:UD-Q5_K_XL
Download
hf download unsloth/DeepSeek-V3.2-GGUF --include "UD-Q5_K_XL/DeepSeek-V3.2-UD-Q5_K_XL-*-of-00010.gguf" --local-dir DeepSeek-V3.2-GGUF
Q6_K513.5 GBfile size, 12 parts514.5 GB513.5 + 0.54 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/DeepSeek-V3.2-GGUF:Q6_K
Download
hf download unsloth/DeepSeek-V3.2-GGUF --include "Q6_K/DeepSeek-V3.2-Q6_K-*-of-00012.gguf" --local-dir DeepSeek-V3.2-GGUF
UD-Q6_K_XL534.6 GBfile size, 12 parts535.6 GB534.6 + 0.54 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/DeepSeek-V3.2-GGUF:UD-Q6_K_XL
Download
hf download unsloth/DeepSeek-V3.2-GGUF --include "UD-Q6_K_XL/DeepSeek-V3.2-UD-Q6_K_XL-*-of-00012.gguf" --local-dir DeepSeek-V3.2-GGUF
Q8_0664.3 GBfile size, 15 parts665.3 GB664.3 + 0.54 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/DeepSeek-V3.2-GGUF:Q8_0
Download
hf download unsloth/DeepSeek-V3.2-GGUF --include "Q8_0/DeepSeek-V3.2-Q8_0-*-of-00015.gguf" --local-dir DeepSeek-V3.2-GGUF
UD-Q8_K_XL726.7 GBfile size, 17 parts727.7 GB726.7 + 0.54 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/DeepSeek-V3.2-GGUF:UD-Q8_K_XL
Download
hf download unsloth/DeepSeek-V3.2-GGUF --include "UD-Q8_K_XL/DeepSeek-V3.2-UD-Q8_K_XL-*-of-00017.gguf" --local-dir DeepSeek-V3.2-GGUF
BF161250.1 GBfile size, 30 parts1251.1 GB1250.1 + 0.54 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/DeepSeek-V3.2-GGUF:BF16
Download
hf download unsloth/DeepSeek-V3.2-GGUF --include "BF16/DeepSeek-V3.2-BF16-*-of-00030.gguf" --local-dir DeepSeek-V3.2-GGUF

MLX for Apple silicon

FormatSizeMemory neededRepo
MLXbaa-ai/DeepSeek-V3.2-RAM-350GB-MLX350.5 GBfile size351.5 GB350.5 + 0.54 + 0.50Download page
MLX 5bitmlx-community/DeepSeek-V3.2-mlx-5bit430.3 GBfile size431.4 GB430.3 + 0.54 + 0.50Download page

Original weights size from the safetensors header, for vLLM, SGLang, Transformers

FormatSizeMemory neededRepo
F8_E4M3original weights642.0 GBfile size643.0 GB642.0 + 0.54 + 0.50
hf download deepseek-ai/DeepSeek-V3.2 --local-dir DeepSeek-V3.2
Other conversions (10; sizes not read)

The Hub tags 21 quantised repos as conversions of this model; we read the most downloaded and show the maker's own first.

Run it by API

Hugging Face Inference Providers, fetched 19 Nov 2026, 13:58 UTC

3 providers serve this model through the Hugging Face router. Prices are US dollars per million tokens. Precision is what the provider reports; a lower precision than the original weights can change answers. Live first, then cheapest output.

ProviderInOutContextPrecisionTokens/sFirst tokenUptime 24hToolsJSON
DeepInfra$0.26$0.38160kfp4250.7 s97.0%yesyes
SiliconFlow$0.26$0.80160kfp8172.2 s98.1%yesyes
DigitalOcean$0.30$0.96160knot given83.2 s98.8%nono

Through the router, with a Hugging Face token (the router bills your Hugging Face account). Change the name after the colon to pick another provider.

curl https://router.huggingface.co/v1/chat/completions \ -H "Authorization: Bearer $HF_TOKEN" -H "Content-Type: application/json" \ -d '{"model": "deepseek-ai/DeepSeek-V3.2:deepinfra", "messages": [{"role": "user", "content": "Hello"}]}'

Not built yet: one key for every open model, with BlockRun, and our own tested-quality score per provider. Until then, precision and measured speed are the best guide to a provider here.

Fine-tunes of this model

from our snapshot of the Hub

Repos that declare this model as their base, most downloaded first. Listing is not a recommendation; unreviewed ones have not been checked.

Planned for this page Not built yet

Try it in private chat
A link to Solana AI's confidential chat for the models it serves. This model is not listed as served there.
Fine-tune it
Fine-tuning a few base models from this page and publishing the result here.
Creators and earnings
Pages for people who publish fine-tunes and adapters of this model, and what they earn.
Fund and bounties
Bounties for new open models, paid from a public fund. How the fund will work.
Contribute compute
How to give compute to open model work, with Akash.

Model data from the Hugging Face Hub 19 Nov 2026, 13:43 UTC. The repo. Found a mistake on this page? Tell us.