The Model Index
Models/Labs/DeepSeek
Lab · 24 models · since Nov 2023

DeepSeek

DeepSeek has 24 models on The Model Index, from DeepSeek Coder (Nov 2023) to DeepSeek-V4.1-Flash (10 Sep 2026). Its strongest is DeepSeek-V4-Pro, with an index score of 167 (GPT-4 = 100), rank 25 of 229. 23 have open weights.

24models tracked23 open · 1 closed
167index score of DeepSeek-V4-Pro#25 of 229 models
10 Sep 2026latest releaseDeepSeek-V4.1-Flash
15current versions9 replaced by newer ones

Every modelnewest first

ModelReleasedWeightsIndex scoreInput $/MOutput $/MContext
31DeepSeek-V4.1-Flash10 Sep 2026Open
165
$0.015$0.401.05M
DeepSeek-V4-Flash-Vision-Exp21 Aug 2026Open
not scored yet
$0.22$0.651.05M
25DeepSeek-V4-Pro13 Aug 2026Open
167
$1.32$3.961.05M
33DeepSeek-V4-Flashreplaced31 Jul 2026Open
165
$0.017$1.281.05M
69DeepSeek-V4-Flash (Apr 2026)replaced24 Apr 2026Open
145
$0.042$0.0841.05M
57DeepSeek-V4-Pro (Apr 2026)replaced24 Apr 2026Open
151
$0.24$0.491.05M
70DeepSeek-V3.21 Dec 2025Open
145
$0.28$0.42164K
DeepSeekMath-V227 Nov 2025Open
not scored yet
–––
80DeepSeek-V3.2-Exp29 Sep 2025Open
143
$0.27$0.41164K
111DeepSeek-V3.1replaced21 Aug 2025Open
132
$0.25$0.95164K
105DeepSeek-R1-052828 May 2025Open
134
–––
133DeepSeek-V3 (Mar 2025)replaced24 Mar 2025Open
123
$0.29$1.14164K
120DeepSeek-R1replaced20 Jan 2025Open
129
–––
127DeepSeek-R1-Distill-Llama-70B20 Jan 2025Open
126
–––
163DeepSeek-R1-Distill-Qwen-1.5B20 Jan 2025Open
108
–––
132DeepSeek-R1-Distill-Qwen-14B20 Jan 2025Open
124
–––
125DeepSeek-R1-Distill-Qwen-32B20 Jan 2025Open
127
–––
145DeepSeek-V3replaced26 Dec 2024Open
116
$0.26$1.03164K
DeepSeek-R1-Lite-Preview20 Nov 2024Closed
not scored yet
–––
DeepSeek-V2.5 (Sep 2024)replaced6 Sep 2024Open
not scored yet
–––
DeepSeek-Coder-V217 Jun 2024Open
not scored yet
–––
DeepSeek-V2replaced6 May 2024Open
not scored yet
–––
214DeepSeek LLM 67B29 Nov 2023Open
82
–––
DeepSeek Coder2 Nov 2023Open
not scored yet
–––

Index score is fitted from independent evaluations and scaled so GPT-4 = 100; a striped bar is an estimate for a model not yet independently tested. Prices are list prices per million tokens.

Release history17 announcements

The Race →
DateReleaseKindSource
10 Sep 2026DeepSeek-V4.1-FlashOpen weightsapi-docs.deepseek.com ↗
21 Aug 2026DeepSeek-V4-Flash-Vision-Exp (weights 08-31)Open weightshuggingface.co ↗
13 Aug 2026DeepSeek-V4-Pro (GA)Flagshipapi-docs.deepseek.com ↗
31 Jul 2026DeepSeek-V4-Flash (official release; 0731 weights)Open weightsapi-docs.deepseek.com ↗
24 Apr 2026DeepSeek-V4 preview (V4-Pro/V4-Flash)Open weightsno link
1 Dec 2025DeepSeek-V3.2 (+ Speciale)Open weightsapi-docs.deepseek.com ↗
27 Nov 2025DeepSeekMath-V2Open weightsno link
29 Sep 2025DeepSeek-V3.2-Exp (sparse attention)Open weightsapi-docs.deepseek.com ↗
21 Aug 2025DeepSeek-V3.1 (hybrid reasoning)Open weightsapi-docs.deepseek.com ↗
28 May 2025DeepSeek-R1-0528Reasoningapi-docs.deepseek.com ↗
20 Jan 2025DeepSeek-R1Reasoningapi-docs.deepseek.com ↗
26 Dec 2024DeepSeek-V3Open weightsapi-docs.deepseek.com ↗
20 Nov 2024DeepSeek-R1-Lite-PreviewReasoningdeepseek.com ↗
17 Jun 2024DeepSeek-Coder-V2Open weightsapi-docs.deepseek.com ↗
6 May 2024DeepSeek-V2Open weightsgithub.com ↗
29 Nov 2023DeepSeek LLM 67BOpen weightsno link
2 Nov 2023DeepSeek CoderOpen weightsno link

Dates come from each lab's own announcement, linked where one exists.