Skip to content
AI Primer

DeepSeek-V3

Mixture-of-experts language model release

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 a 128K context length.

Pricing

Official site · Aug 9, 2026, 7:11 AM
Input / 1M
$0.27
Output / 1M
$1.10
Cached input / 1M
$0.07

Per 1M tokens. Input price is for cache miss; cached input/cache hit price is $0.07 per 1M tokens.

DeepSeek's first-party DeepSeek-V3 announcement/API docs page explicitly states the DeepSeek-V3 API pricing from Feb 8 onward as $0.27 per 1M input tokens on cache miss, $0.07 per 1M input tokens on cache hit, and $1.10 per 1M output tokens. The current general Models & Pricing page read on 2026-08-09 lists only DeepSeek V4 models, so this is the latest exact DeepSeek-V3-specific first-party numeric pricing found rather than a current V4 price inferred for V3.

View source

Model Intelligence

Context window
128,000 tokens
Arena ranking
14
Benchmarkable
Yes
Model level
release
Intelligence Index
14.2
Coding Index
23
Math Index
26
MMLU Pro
0.75
GPQA
0.56
HLE
0.03
LiveCodeBench
0.36
SciCode
0.35
MATH-500
0.89
AIME
0.25
AIME 2025
0.26
IFBench
0.35
LCR
0.32
TerminalBench Hard
0.07
TAU2
0.23

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.