← Back to AI Models

DeepSeek V3

MoE model with 671B parameters achieving frontier performance

deepseekai-modelmigration

Overview

MoE model with 671B parameters achieving frontier performance The model has 671B (37B active) parameters. It is an open-weight model released under the MIT licence.

Key Capabilities

  • Code Generation — suited for tasks requiring code generation proficiency
  • Reasoning — suited for tasks requiring reasoning proficiency
  • Math — suited for tasks requiring math proficiency
  • Multilingual — suited for tasks requiring multilingual proficiency

Recommended Use Cases

  • Cost Effective Migration
  • Large Scale Processing
  • Code Analysis

Technical Specifications

  • Context Window: 128K tokens
  • Parameters: 671B (37B active)
  • Open Weight: Yes
  • Licence: MIT
  • Release Date: December 25, 2024

Pricing

$0.27/M input tokens, $1.1/M output tokens

Provider

DeepSeek V3 is developed by DeepSeek. For the latest information, visit the official documentation.