DeepSeek: DeepSeek V3

DeepSeek: DeepSeek V3

deepseek · Released Dec 26, 2024 Efficient
Intelligence #162 / 525
45.4 Our Score
AA Index #187 / 353
16.5 Artificial Analysis
Input #317 / 525
$0.320 per 1M tokens
Output #288 / 525
$0.890 per 1M tokens
Context #174 / 525
163,840 tokens

Analysis Summary

DeepSeek: DeepSeek V3 sits in the Efficient tier on our leaderboard, ranked #162 of 525 published models on overall intelligence. At $0.320 input and $0.890 output per 1M tokens, it is among the most expensive on the market. It offers a standard large context window and supports tool use and function calling.

Editorial notes

DeepSeek V3 is a highly cost-effective model with strong coding and maths performance relative to its price, making it one of the better value propositions for technical business tasks. Businesses should note that provider accessibility may vary depending on their region and compliance requirements.

Assessed April 23, 2026

Rankings consider pricing, capabilities, benchmarks, and real-world applicability and are refreshed as new models launch. Feedback?

Performance Profile

Intelligence3.5Technical2.8Value7.8Content4.5
Intelligence 3.5/10
Technical 2.8/10
Content 4.5/10
Value 7.8/10

DeepSeek-V3 is the latest model from the DeepSeek team, building upon the instruction following and coding abilities of the previous versions. Pre-trained on nearly 15 trillion tokens, the reported evaluations..

Capabilities

Tool Use Function Calling

Performance Indices

Source: Artificial Analysis

16.5 Intelligence Index
16.4 Coding Index
14.8 Agentic Index
26 Math Index

Benchmark Scores

Intelligence

GPQA Diamond 55.7% Graduate-level scientific reasoning
HLE 3.6% Humanity's Last Exam
MMLU Pro 75.2% Multi-task language understanding
MATH 500 88.7% Mathematical problem-solving
AIME 25.3% Competition mathematics
AIME 2025 26% Competition mathematics (2025)
SciCode 35.4% Scientific computing

Technical

LiveCodeBench 35.9% Live coding evaluation
TerminalBench Hard 6.8% Agentic terminal tasks
τ²-Bench 22.8% Conversational agent benchmark

Content

IFBench 34.8% Instruction following
LCR 29% Long-context reasoning

Benchmark data from Artificial Analysis and Hugging Face

How does DeepSeek: DeepSeek V3 stack up?

Compare side-by-side with other efficient models.

Compare Models

Model Information

OpenRouter ID deepseek/deepseek-chat
Providerdeepseek
Model FamilyDeepSeek
Release Date December 26, 2024
Context Length163,840 tokens
Max Completion16,384 tokens
Status Active

Pricing

Token Type Cost per 1M tokens Cost per 1K tokens
Input $0.32 $0.000320
Output $0.89 $0.000890

Live Performance

Live endpoint metrics — refreshed every 30 minutes.

98.8%
Avg Uptime
1,218ms
Best Latency (TTFT)
8 tok/s
Best Throughput
2/2
Active Endpoints
Available via: DeepInfra, Novita

Leaderboard Categories