Skip to content
AI Primer

DeepSeek-V3

A strong Mixture-of-Experts (MoE) language model.

DeepSeek-V3 is a specific open-source Mixture-of-Experts language model release from DeepSeek, with 671B total parameters, 37B activated parameters per token, trained on 14.8T tokens, and released with model weights, technical report, chat/API access, and local inference guidance.

Pricing

Official site · Aug 10, 2026, 1:11 PM
Input / 1M
$0.27
Output / 1M
$1.10
Cached input / 1M
$0.07

First-party DeepSeek-V3 announcement states these rates apply 'From Feb 8 onwards.' Current DeepSeek Models & Pricing page says billing is per 1M input/output tokens and currently lists newer V4 models, not V3.

DeepSeek's first-party DeepSeek-V3 announcement explicitly lists API pricing from Feb. 8 onward at $0.27 per 1M input tokens for cache misses, $0.07 per 1M cached input tokens, and $1.10 per 1M output tokens. The current first-party Models & Pricing page confirms DeepSeek prices are billed per 1M tokens but currently lists newer V4 models rather than DeepSeek-V3, so confidence is medium for current availability while the V3 numeric pricing is directly cited.

View source

Model Intelligence

Context window
128,000 tokens
Arena ranking
15
Benchmarkable
Yes
Model level
release
Intelligence Index
15.2
Coding Index
21.2
Math Index
41
MMLU Pro
0.82
GPQA
0.66
HLE
0.05
LiveCodeBench
0.41
SciCode
0.36
MATH-500
0.94
AIME
0.52
AIME 2025
0.41
IFBench
0.41
LCR
0.41
TerminalBench Hard
0.15
TAU2
0.47

Recent stories

0 linked stories
No linked stories yet.
AI PrimerAI Primer

Your daily guide to AI tools, workflows, and creative inspiration.

© 2026 AI Primer. All rights reserved.