DeepSeek V3 0324

Creator

DeepSeek

Released

2025-03-25

Intelligence

15.4

Artificial Analysis Index

Coding

21.2

Artificial Analysis Index

In $/1M

$0.27

input tokens

Out $/1M

$1.12

output tokens

Blended $/1M

$0.48

3:1 blended

Speed

0

tokens / sec

Profile

License, openness, and modality — the governance layer.

License

MIT + Model License (Commercial use allowed)

non-commercial

Weights

Open

downloadable

Modalities

Text

Parameters

671B

total

A powerful Mixture-of-Experts (MoE) language model with 671B total parameters (37B activated per token). Features Multi-head Latent Attention (MLA), auxiliary-loss-free load balancing, and multi-token prediction training. Pre-trained on 14.8T tokens with strong performance in reasoning, math, and code tasks.

Capability profile

Category strength across 20 domains, via LLM Stats.

math
80legal
80finance
80language
80healthcare
80biology
70
general
70physics
70chemistry
70reasoning
70code
50
Via LLM Stats

Benchmark breakdown

Independent evaluation scores, normalized to 0–100.

GPQA Diamond
66
Humanity’s Last Exam
5
MMLU-Pro
SciCode
36
LiveCodeBench
41
MATH-500
94
AIME 2025
41
τ²-Bench (agentic)
47
Terminal-Bench Hard
15
IFBench
41
Via Artificial Analysis
© 2026 NYSGPT2525 LLC