DeepSeek V3 0324

Mar 2025
DeepSeek V3 0324
Compare in
on
$0.021per task?
$0.24 / $0.90per 1M in / out?
97 tok/sfirst token in 0.7 s?
128Kcontext

Rankingin non-thinking mode?

Strong at

B
Tool calling#113 of 242
C
Coding#75 of 132Agents & work tasks#122 of 175Reasoning#287 of 334Science#211 of 312Long context#196 of 273Knowledge#175 of 318Instructions#154 of 245

Alternatives?

ModelScoreCostPer taskWhat changes
CheaperAgnes 2.5 Pro Beta×3.6-63%$0.10 / $0.30BetterGPT-6 Lunalow×2.2-79%$0.005Function calling only with none · Chat Completions+3OpenMiMo v2.6 Prothinking×4.8+530%$0.13Memory (4-bit): 411 GB → 615 GB+4CheaperGPT-5.6 Lunamedium×2.6-26%$0.016CheaperLing 3.0 Flash VLthinkingOpen×2.5-78%$0.06 / $0.18

Run it yourself

estimate
Fits on✗ 8 GB (phone, laptop)✗ 16 GB✗ 24 GB GPU✗ 48 GB GPU✗ 80 GB GPU
LicenceMITfree for commercial use

History

5 versions
101520Jan 25Apr 25Jul 25Oct 25Release dateAA Intelligence Index 4.3.2V3 · DecV3 · MarV3.1V3.2
101520Jan 25Jul 25Release dateAA Intelligence Index 4.3.2V3 · DecV3 · MarV3.1V3.2

Benchmarks

all categories
Chat & Knowledge
Humanity's Last Exam4.7#289
AA Omniscience Index-40.7#176
CritPt (physics)0#180
AA Intelligence Index 4.3.29.7#212
AA Omniscience · Accuracy24.3#121
AA Omniscience · Hallucination Rate85.9#211
AA-LCR (long context) 1.140.7#198
GDP.pdf · AA all-pass1.6#109
MMLU-Pro · official boardlab81.3
GPQA Diamond65.5#213
AIME 2025 · AA archived41#95
Coding
Terminal-Bench 4.00#75
SciCode39#99
Terminal-Bench Hard (version unstated)15.2#109
Terminal-Bench 2.113.9#119
LiveCodeBench · AA archived40.5#104
Agents & Tool Use
AutomationBench-AA0.3#127
GDPval-AA (work tasks) 2.10#122
τ³-Banking · AA4.7#123
AA-Briefcase 1.1 · Elo68.1#126
τ²-bench · AA47.1#115
Writing & Language
IFBench (instructions)41#155

FAQ

How much does DeepSeek V3 0324 cost?›

$0.24 per 1M input tokens and $0.90 per 1M output tokens (cheapest host: DeepInfra). A typical benchmark task costs about $0.021 at default effort.

Is DeepSeek V3 0324 free to use?›

There is no free API; the cheapest paid price is $0.24 per 1M input tokens. The weights are open, so you can run it yourself.

How good is DeepSeek V3 0324, and what is it best at?›

#287 of 334 in Chat & Knowledge, #75 of 132 in Coding, #127 of 133 in Agents & Tool Use, #85 of 187 in Writing & Language, #109 of 133 in Vision & Documents.

Is DeepSeek V3 0324 better than DeepSeek R1 0528 Qwen3 8B?›

On AA Intelligence Index 4.3.2, DeepSeek V3 0324 scores 9.7 vs 8.1 — higher.

How can I use DeepSeek V3 0324?›

Through the API of DeepInfra, Featherless, GMICloud, SiliconFlow.

What is the context window of DeepSeek V3 0324?›

128K tokens, with up to 115K output tokens.

When was DeepSeek V3 0324 released? Is there a newer version?›

Released 2025-03-25. It has been superseded by DeepSeek V3.1.

Is DeepSeek V3 0324 open source?›

Yes, the weights are open (MIT licence).

Can I run DeepSeek V3 0324 locally?›

The weights are open and can be self-hosted; hardware needs are not confirmed yet.

Can DeepSeek V3 0324 read images?›

No, it accepts text only.

What is DeepSeek V3 0324?›

DeepSeek V3 0324 is an open-weights language model from DeepSeek, released Mar 2025, the successor to DeepSeek R1 0528 Qwen3 8B.

What are the best alternatives to DeepSeek V3 0324?›

Agnes 2.5 Pro Beta (cheaper), GPT-6 Luna (better), MiMo v2.6 Pro (open weights).