NeuralTimes

DeepSeek V3

China's Open Source AI Giant

Tap to continue
NeuralTimes
1 / 9

What is DeepSeek?

Chinese AI startup
Open source LLM

Rivals GPT-4!
Completely free

NeuralTimes
2 / 9

The Numbers

685B parameters
60T training tokens

One of the largest
open source models!

NeuralTimes
3 / 9

Benchmark Scores

MMLU: 87.1%
MATH: 90.2%
Code: 82.6%

Beats many rivals!

NeuralTimes
4 / 9

The Secret

Mixture of Experts

Only 37B active params
Very efficient
Fast inference

NeuralTimes
5 / 9

Training Cost

$5.5M only!

vs $100M+ for GPT-4

20x cheaper! 🤯

NeuralTimes
6 / 9

Why It Matters

Proves:
→ Don't need billions
→ Open source works
→ Competition good

NeuralTimes
7 / 9

How to Use

Download: HuggingFace
Run: vLLM, Ollama
API: api.deepseek.com

Try it free!

📖

Full Analysis

Read on NeuralTimes

Read Full Article