Skip to content
Weights Not Walls
Menu

deepseek-ai

DeepSeek-V4-Pro

Reviewed
Size
1.6T 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
22 Apr 2026the date the repo was created
Context length
1M tokensfrom the model's config
Kind
Base model (no base model declared on the Hub)DeepseekV4ForCausalLM
On Hugging Face
deepseek-ai/DeepSeek-V4-Pro321k downloads in the last 30 days, 5.7k likes

Scores

24 independent results on 16 eval sets
Chess PuzzlesRe-run by Epoch AI
  • effort max19.0%±3.913 Oct 2026, log
  • effort high16.0%±3.713 Oct 2026, log
  • effort none4.0%±2.013 Oct 2026, log

Epoch's name: deepseek-v4-pro. Best closed model on this set: claude-fable-5-2_max, 79.0% (Anthropic). All 166 results

  • effort max2.4%±2.417 Jun 2026, log

Epoch's name: deepseek-v4-pro. Best closed model on this set: claude-fable-5-2_max, 100.0% (Anthropic). All 83 results

  • effort max45.3%±3.017 Jun 2026, log

Epoch's name: deepseek-v4-pro. Best closed model on this set: claude-fable-5-2_max, 95.1% (Anthropic). All 127 results

GPQA diamondRe-run by Epoch AI
  • effort high90.9%±2.06 Aug 2026, log
  • effort max89.6%±1.716 Jun 2026, log
  • effort none73.2%±3.26 Aug 2026, log

Epoch's name: deepseek-v4-pro. Best closed model on this set: gpt-6.2-astra_max, 96.8% (OpenAI). All 341 results

Mystery Game PuzzlesRe-run by Epoch AI
  • effort none17.0%±3.827 Aug 2026, log

Epoch's name: deepseek-v4-pro. Best closed model on this set: gpt-6.2-astra_max, 95.0% (OpenAI). All 148 results

OTIS Mock AIME 2024-2025Re-run by Epoch AI
  • effort max96.7%±2.017 Jun 2026, log
  • effort high95.6%±3.16 Aug 2026, log
  • effort none46.7%±7.56 Aug 2026, log

Epoch's name: deepseek-v4-pro. Best closed model on this set: claude-fable-5_high, 100.0% (Anthropic). All 318 results

SimpleQA VerifiedRe-run by Epoch AI
  • effort max47.0%±1.627 Aug 2026, log

Epoch's name: deepseek-v4-pro. Best closed model on this set: claude-fable-5-2_max, 94.0% (Anthropic). All 105 results

SWE-Bench verifiedRe-run by Epoch AI
  • effort max77.6%±1.918 Jun 2026, log

Epoch's name: deepseek-v4-pro. Best closed model on this set: claude-opus-4-7_max, 83.5% (Anthropic). All 35 results

Epoch's name: deepseek-v4-pro. Best closed model on this set: gpt-5.5_high, 22.2% (OpenAI). All 17 results

CritPtReported

Epoch's name: deepseek-v4-pro. Best closed model on this set: gemini-4-carbon_high, 36.0% (Google DeepMind). All 271 results

DTBenchReported

Epoch's name: deepseek-v4-pro. Best closed model on this set: claude-fable-5-2_unknown, 99.2% (Anthropic). All 241 results

FrontierCodeReported
  • effort none; Harness: mini-swe-agent; Reasoning effort: none17.6%cognition.com

Epoch's name: deepseek-v4-pro. Best closed model on this set: claude-opus-5-5_medium, 54.6% (Anthropic). All 46 results

LMCAReported

Epoch's name: deepseek-v4-pro. Best closed model on this set: claude-fable-5-2_xhigh, 70.6% (Anthropic). All 203 results

ProofBenchReported
  • effort max; Reasoning effort: max16.0%±3.7vals.ai

Epoch's name: deepseek-v4-pro. Best closed model on this set: claude-fable-5-1_max, 100.0% (Anthropic). All 85 results

Epoch's name: deepseek-v4-pro. Best closed model on this set: claude-fable-5_high, 95.0% (Anthropic). All 29 results

WeirdMLReported

Epoch's name: deepseek-v4-pro. Best closed model on this set: gpt-6-astra_promax, 93.6% (OpenAI). All 173 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: 0.95 GB, worked out from the model's config (deepseek-ai/DeepSeek-V4-Pro), 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

teamblobfish/DeepSeek-V4-Pro-GGUF (converted by teamblobfish), 620 downloads. Sizes are the real files on the Hub, all parts added up.

FormatSizeMemory neededGet it
Q2_K534.7 GBfile size, 13 parts536.2 GB534.7 + 0.95 + 0.50
Commands
llama.cpp
llama-server -hf teamblobfish/DeepSeek-V4-Pro-GGUF:Q2_K
Download
hf download teamblobfish/DeepSeek-V4-Pro-GGUF --include "Q2_K-XL/DeepSeek-V4-Pro-Q2_K-XL-*-of-00013.gguf" --local-dir DeepSeek-V4-Pro-GGUF
Q4_K_M889.0 GBfile size, 21 parts890.5 GB889.0 + 0.95 + 0.50
Commands
llama.cpp
llama-server -hf teamblobfish/DeepSeek-V4-Pro-GGUF:Q4_K_M
Download
hf download teamblobfish/DeepSeek-V4-Pro-GGUF --include "Q4_K_M-XL/DeepSeek-V4-Pro-Q4_K_M-XL-*-of-00021.gguf" --local-dir DeepSeek-V4-Pro-GGUF
Q8_01557.0 GBfile size, 37 parts1558.5 GB1557.0 + 0.95 + 0.50
Commands
llama.cpp
llama-server -hf teamblobfish/DeepSeek-V4-Pro-GGUF:Q8_0
Download
hf download teamblobfish/DeepSeek-V4-Pro-GGUF --include "Q8_0/DeepSeek-V4-Pro-Q8_0-*-of-00037.gguf" --local-dir DeepSeek-V4-Pro-GGUF

MLX for Apple silicon

FormatSizeMemory neededRepo
MLXinferencerlabs/DeepSeek-V4-Pro-MTP-MLX13.0 GBfile size14.5 GB13.0 + 0.95 + 0.50Download page
MLXmlx-community/DeepSeek-V4-Pro791.3 GBfile size792.8 GB791.3 + 0.95 + 0.50Download page
Other conversions (13; sizes not read)

The Hub tags 31 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

8 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
Relace$0.90$1.501Mfp4940.9 s99.9%yesno
SiliconFlow$0.90$1.701Mfp8302.3 s99.9%yesno
DigitalOcean$1.04$2.091Mnot given401.1 s100.0%yesno
Novita$1.28$2.561Mfp8602.2 s100.0%yesno
DeepInfra$1.30$2.601Mfp8431.5 s99.9%yesyes
Venice$1.65$3.30977knot given571.8 s95.2%yesyes
Azure$1.91$3.831Mnot given573.3 s100.0%yesno
featherless-ai???not given???not givennot given

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-V4-Pro:relace", "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.

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.