Skip to main content
DeepSeek·current

DeepSeek V3

DeepSeek's efficient MoE model — GPT-4-class performance at a fraction of the training cost

ProviderDeepSeek
Released2024-12-26
StatusCurrent
Context window66K tokens
PricingOpen
Input price$0.27/M
Output price$1.10/M
Capabilities
textcodefunction-calling
Hugging FaceView on HF ↗

Benchmarks

MMLU (knowledge)88.5
HumanEval (coding)91.6
MATH (mathematics)90.2

As of 2025-02-01. Source

Loading…

Compare with similar models

DeepSeek V3Llama 3.1 405BDeepSeek R1Llama 3.3 70B
Context66K131K66K131K
MMLU88.588.690.886.0
HumanEval91.6
MATH90.273.597.377.0
GPQA71.5
PricingOpenOpenOpenOpen
Input $/M$0.27$0.55

Compare more models →