DeepSeek: DeepSeek-V3.2 API
DeepSeek-V3.2 is the definitive "Reasoning-First" multimodal foundation model, utilizing the third-generation Multi-head Latent Attention (MLA) and DeepSeek-MoE architecture. This version introduces the "Dynamic Token Pruning" technology, which reduces inference latency by 40% compared to V3.0 while maintaining top-tier coding and mathematical reasoning capabilities. V3.2 is natively multimodal, capable of processing interleaved text, high-resolution images, and long-form video inputs without separate vision encoders. In 2026, it is widely recognized as the most cost-effective "GPT-5 Class" model, offering open-source weights for researchers and a highly scalable API for global developers.
- Context window: 128,000 tokens
- Max output: 64,000 tokens
- Input: text, image
- Output: text
- Reasoning: Supported
- Tool calling: Supported
- File input: Supported
- Released: 2025-11-30
- Knowledge cutoff: 2024-06-30
Frequently Asked Questions
What is the context window of DeepSeek-V3.2?
DeepSeek-V3.2 supports a context window of up to 128,000 tokens.
Does DeepSeek-V3.2 support function calling?
Yes. DeepSeek-V3.2 supports tool / function calling.
Does DeepSeek-V3.2 support reasoning?
Yes. DeepSeek-V3.2 is a reasoning-capable model.
What is the knowledge cutoff of DeepSeek-V3.2?
The knowledge cutoff of DeepSeek-V3.2 is 2024-06-30.