Benchmark Almanac

DeepSeek

Best on the Capabilities Index DeepSeek V4.1 Flash 1017 158.8
38 models 20 model cards read deepseek.com

DeepSeek's models against the field

Every model on the Epoch Capabilities Index by release date, with DeepSeek's in color.

50 75 100 125 150 175 2023 2024 2025 2026 2027 DeepSeek V4.1 Flash 1017 DeepSeek-V4-Flash 50 75 100 125 150 175 2023 2024 2025 2026 2027 DeepSeek V4.1 Flash 1017 DeepSeek-V4-Flash

Its strongest models

DeepSeek Oct 2026

DeepSeek V4.1 Flash 1017

Open weights Mid-size

158.8 #19 of 271
Epoch Capabilities Index
  • NL2Repo-Bench 68.2% #1 /28
  • MultiChallenge 74.9% #5 /35
  • Harvey LAB (Legal Agent Benchmark) 9.17% #24 /80
DeepSeek Sep 2026

DeepSeek V4.1 Flash

Open weights Mid-size

156.1 #34 of 271
Epoch Capabilities Index
  • MMMU-Pro 56.5% #61 /72
  • NL2Repo-Bench 65.4% #2 /28
  • LongBench v2 45.2% #24 /27
DeepSeek Aug 2026

DeepSeek V4 Pro 0813

Open weights Flagship

155.9 #35 of 271
Epoch Capabilities Index
  • NL2Repo-Bench 61.5% #3 /28
  • Harvey LAB (Legal Agent Benchmark) 7.50% #28 /80
  • CritPt 18% #35 /160
DeepSeek Jul 2026

DeepSeek V4 Flash 0731

Open weights Mid-size

154.9 #41 of 271
Epoch Capabilities Index
  • NL2Repo-Bench 54.2% #10 /28
  • Harvey LAB (Legal Agent Benchmark) 8.33% #26 /80
  • CritPt 16.6% #42 /160
DeepSeek Apr 2026

DeepSeek-V4-Pro

Open weights Flagship

149.1 #67 of 271
Epoch Capabilities Index
  • SWE-bench Verified 77.6% #6 /32
  • DeepSWE 62.8% #14 /28
  • Toolathlon 52.8% #5 /50
DeepSeek Apr 2026

DeepSeek-V4-Flash

Open weights

146.2 #83 of 271
Epoch Capabilities Index
  • DeepSWE 53.3% #20 /28
  • Toolathlon 48.2% #10 /50
  • Claw-Eval 74.2% #17 /23

What its model cards measure

Tall ticks are DeepSeek model cards that list the benchmark in their headline table. Gray dots are cards that leave it out. Rows are the benchmarks the lab listed most, at least twice each, in the order it started using them.

HumanEval C-Eval MMLU GSM8K BIG-Bench Hard HellaSwag CMMLU LiveCodeBench MATH AlpacaEval 2.0 SWE-bench Verified MMLU-Pro GPQA Diamond Codeforces rating SimpleQA Aider Polyglot MMLU-Redux AIME 2024 DROP Humanity's Last Exam BrowseComp SWE-bench Multilingual Toolathlon NL2Repo-Bench DeepSWE Agents' Last Exam AutomationBench DSBench (DeepSeek internal)

Model cards

DateModelDocumentBenchmarks
17 Oct 2026 DeepSeek V4.1 Flash 1017 DeepSeek-V4.1-Flash-1017 model card 11
10 Sep 2026 DeepSeek V4.1 Flash DeepSeek-V4.1-Flash model card 37
21 Aug 2026 DeepSeek-V4-Flash-Vision-Exp DeepSeek-V4-Flash-Vision-Exp model card (Hugging Face) 10
13 Aug 2026 DeepSeek V4 Pro 0813 DeepSeek-V4-Pro-0813 model card 10
31 Jul 2026 DeepSeek V4 Flash 0731 DeepSeek-V4-Flash-0731 model card 9
24 Apr 2026 DeepSeek-V4-Pro DeepSeek-V4-Pro model card 43
1 Dec 2025 DeepSeek-V3.2 DeepSeek-V3.2 technical report, Table 2 18
27 Nov 2025 DeepSeekMath-V2 DeepSeekMath-V2: Towards Self-Verifiable Mathematical Reasoning (model card) 3
29 Sep 2025 DeepSeek-V3.2-Exp DeepSeek-V3.2-Exp model card 14
22 Sep 2025 DeepSeek-V3.1-Terminus DeepSeek-V3.1-Terminus model card 12
21 Aug 2025 DeepSeek-V3.1 DeepSeek-V3.1 model card 16
28 May 2025 DeepSeek-R1 DeepSeek-R1-0528 model card 16
30 Apr 2025 DeepSeek-Prover-V2-671B DeepSeek-Prover-V2 model card (Hugging Face) 3
24 Mar 2025 DeepSeek-V3 DeepSeek-V3-0324 model card 5
27 Jan 2025 Janus-Pro-7B Janus-Pro: Unified Multimodal Understanding and Generation with Data and Model Scaling (GitHub README and tech report) 9
20 Jan 2025 DeepSeek-R1 DeepSeek-R1 model card 20
26 Dec 2024 DeepSeek-V3 DeepSeek-V3 model card 48
5 Sep 2024 DeepSeek-V2.5 Sep 2024 DeepSeek-V2.5 model card 11
7 May 2024 DeepSeek-V2 (MoE-236B, May 2024) DeepSeek-V2 model card 13
29 Nov 2023 DeepSeek LLM 67B DeepSeek LLM GitHub README 12

Every DeepSeek model