Project AURIONAI Use-Cases Research & Innovation
DeepSeek

DeepSeek

Startup, China. 36 models in the directory.

startupChina
0deployments

Identity

Type
startup
Country label
China
HQ country
China
Founding country
China
Parent
not in the notes
Parent country
not in the notes
Country basis
the research notes
Where this comes from
Convention D1: a dual label, HQ first, when the HQ, founding or parent country differ.

Models by Class

LLM15

Models

ModelClassOpennessParametersReleased
DeepSeekDeepSeek-V4.1-Flashmultimodal LLMopen weights permissive552B10 Sep 2026DeepSeekDeepSeek-V4-Flash Vision-Expmultimodal LLMopen weights permissive–1 Sep 2026DeepSeekDeepSeek-V4-Pro-0813LLMopen weights permissive–13 Aug 2026DeepSeekDeepSeek-V4-Flash-0731LLMopen weights permissive–31 Jul 2026DeepSeekDeepSeek-V4-FlashLLMopen weights permissive284B24 Apr 2026DeepSeekDeepSeek-V4-ProLLMopen weights permissive1.6T24 Apr 2026DeepSeekDeepSeek V4LLMopen weights permissive–22 Apr 2026DeepSeekDeepSeek-V4-Flash BaseLLMopen weights permissive–22 Apr 2026DeepSeekDeepSeek-OCR 2OCR and documentopen weights permissive3.39B27 Jan 2026DeepSeekDeepSeek-V3.2reasoningopen weights permissive–1 Dec 2025DeepSeekDeepSeek-OCROCR and documentopen weights permissive3.34B17 Oct 2025DeepSeekDeepSeek-V3.2-ExpLLMunknown–29 Sep 2025DeepSeekDeepSeek-V3.1-TerminusLLMopen weights permissive–22 Sep 2025DeepSeekDeepSeek-V3.1LLMopen weights permissive671B21 Aug 2025DeepSeekDeepSeek-Prover-V2 7Breasoningunknown6.91B30 Apr 2025DeepSeekJanus-Pro 1Bmultimodal LLMopen weights permissive–29 Jan 2025DeepSeekJanus-Pro 7Bmultimodal LLMopen weights permissive–29 Jan 2025DeepSeekDeepSeek-R1-Distill-Llama-70Breasoningopen weights restricted–20 Jan 2025DeepSeekDeepSeek-R1-Distill-Llama-8Breasoningopen weights restricted–20 Jan 2025DeepSeekDeepSeek-R1-Distill-Qwen-1.5Breasoningopen weights permissive–20 Jan 2025DeepSeekDeepSeek-R1-Distill-Qwen-14Breasoningopen weights permissive–20 Jan 2025DeepSeekDeepSeek-R1-Distill-Qwen-32Breasoningopen weights permissive–20 Jan 2025DeepSeekDeepSeek-R1-Distill-Qwen-7Breasoningopen weights permissive–20 Jan 2025DeepSeekDeepSeek-R1reasoningunknown–Jan 2025DeepSeekDeepSeek-V2.5 (Dec 2024, DeepSeek-V2.5-1210)LLMopen weights restricted236B10 Dec 2024DeepSeekDeepSeek v3LLMunknown671BDec 2024DeepSeekJanusFlow 1.3Bmultimodal LLMopen weights permissive2.05B12 Nov 2024DeepSeekJanus 1.3Bmultimodal LLMopen weights permissive2.09B18 Oct 2024DeepSeekJanusmultimodal LLMopen weights permissive–17 Oct 2024DeepSeekDeepSeek-V2.5LLMopen weights restricted–5 Sep 2024DeepSeekDeepSeek-Coder-V2codeopen weights permissive236B14 Jun 2024DeepSeekDeepSeek-Coder-V2-Basecodeopen weights permissive236B14 Jun 2024DeepSeekDeepSeek-V2LLMunknown–7 May 2024DeepSeekDeepSeekMath 7B Instructreasoningopen weights restricted–5 Feb 2024DeepSeekdeepseek-llm-67b-baseLLMunknown67B29 Nov 2023DeepSeekDeepSeek LLM 67B Chat (V1)LLMopen weights permissive67B29 Nov 2023

Scores

BenchmarkModelScoreWho ran itSetting
AIME 2025DeepSeek-V3.188.4 %vendor-reportedThinking mode, Pass@1
Arena Text leaderboardDeepSeek-V4-Pro-08131,465 Eloindependentdeepseek-v4-pro-high-20260813
Arena Text leaderboardDeepSeek-V4.1-Flash1,475 Eloindependentdeepseek-v4.1-flash-max
Artificial Analysis Intelligence IndexDeepSeek-V4.1-Flash39 indexindependentmax
BrowseCompDeepSeek-V3.130 %vendor-reportedThinking mode, internal search agent framework
BrowseCompDeepSeek-V3.1-Terminus38.5 %vendor-reportedModel card comparison table; mode not stated
Codeforces ratingDeepSeek-V4-Pro3,206 ratingvendor-reported
Codeforces ratingDeepSeek-V4.1-Flash3,471 ratingvendor-reported
DeepSWEDeepSeek-V4.1-Flash74.2 %vendor-reportedmini-SWE
GPQA DiamondDeepSeek-V3.180.1 %vendor-reportedThinking mode, Pass@1
GPQA DiamondDeepSeek-V3.1-Terminus80.7 %vendor-reportedModel card comparison table; mode not stated
GPQA DiamondDeepSeek-V4-Flash89.9 %vendor-reported
GPQA DiamondDeepSeek-V4-Pro90.1 %vendor-reported
GPQA DiamondDeepSeek-V4.1-Flash90.9 %vendor-reported
Humanity's Last Exam (HLE)DeepSeek-V4-Pro37.7 %vendor-reported
LiveCodeBenchDeepSeek-V2.5 (Dec 2024, DeepSeek-V2.5-1210)34.38 %vendor-reported2024-08-01 to 2024-12-01 window; up from 29.2
LiveCodeBenchDeepSeek-V3.174.8 %vendor-reportedThinking mode, 2408 to 2505, Pass@1
LiveCodeBenchDeepSeek-V3.1-Terminus74.9 %vendor-reportedModel card comparison table; mode not stated
LiveCodeBenchDeepSeek-V4-Flash91.6 %vendor-reported
LiveCodeBenchDeepSeek-V4-Pro93.5 %vendor-reported
MATH500DeepSeek-V2.5 (Dec 2024, DeepSeek-V2.5-1210)82.8 %vendor-reportedUp from 74.8 for DeepSeek-V2.5
MMLU-ProDeepSeek-V3.184.8 %vendor-reportedThinking mode, EM
MMLU-ProDeepSeek-V3.1-Terminus85 %vendor-reportedModel card comparison table; mode not stated
MMLU-ProDeepSeek-V4-Pro87.5 %vendor-reported
OmniDocBenchDeepSeek-OCR 290.25 scorevendor-reportedoverall
Real5-OmniDocBenchDeepSeek-OCR73.99 scorevendor-reportedoverall
Real5-OmniDocBenchDeepSeek-OCR 273.01 scorevendor-reportedoverall
SWE-bench MultilingualDeepSeek-V3.154.5 %vendor-reportedNon-thinking, agent mode
SWE-bench MultilingualDeepSeek-V3.1-Terminus57.8 %vendor-reportedModel card comparison table; mode not stated
SWE-bench VerifiedDeepSeek-V3.166 %vendor-reportedNon-thinking, agent mode
SWE-bench VerifiedDeepSeek-V3.1-Terminus68.4 %vendor-reportedModel card comparison table; mode not stated
SWE-bench VerifiedDeepSeek-V4-Flash79 %vendor-reported
SWE-bench VerifiedDeepSeek-V4-Pro80.6 %vendor-reported
Terminal-BenchDeepSeek-V4-Flash56.9 %vendor-reported
Terminal-BenchDeepSeek-V4-Pro67.9 %vendor-reported
Terminal-BenchDeepSeek-V4.1-Flash90.6 %vendor-reported
Terminal-BenchDeepSeek-V4.1-Flash90.6 %vendor-reportedDeepSeek Harness Minimal
Terminal-BenchDeepSeek-V4.1-Flash30vendor-reported
Terminal-BenchDeepSeek-V4.1-Flash31.2 %vendor-reported
Where this comes from
Every recorded score of this organisation's models, with who ran it. Scores from different settings are not directly comparable.

Sources

  1. LLM Leaderboard - Comparison of AI models from OpenAI, Anthropic, Google, SpaceXAI & others | Artificial Analysis artificialanalysis.ai https://artificialanalysis.ai/leaderboards/models
  2. deepseek-ai (DeepSeek) huggingface.co https://huggingface.co/deepseek-ai
  3. arxiv.org/pdf/2606.03264 https://arxiv.org/pdf/2606.03264
  4. deepseek-ai/DeepSeek-R1-Distill-Qwen-32B , Hugging Face huggingface.co https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B
  5. DeepSeek-V3 Technical Report arxiv.org https://arxiv.org/html/2412.19437v1
  6. DeepSeek on the Cover of "Nature": First Response to Controversies Led by LIANG Wenfeng eu.36kr.com https://eu.36kr.com/en/p/3471852231792260
  7. DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning arxiv.org https://arxiv.org/abs/2501.12948
  8. Introducing DeepSeek-V3.2-Exp | DeepSeek API Docs api-docs.deepseek.com https://api-docs.deepseek.com/news/news250929

To cite this page: Project AURION, AI Use-Cases Research & Innovation, "DeepSeek", https://aurion.semai.id/organisations/deepseek/, records of October 2026.