DeepSeek: deepseek-reasoner API
deepseek-reasoner is the specialized reasoning API alias from DeepSeek, designed to tackle complex logic, mathematics, programming, and long-horizon planning. Unlike the conversation-oriented deepseek-chat, it mandates the generation of a Chain of Thought (CoT) before delivering the final answer. As of early 2026, this endpoint primarily powers the DeepSeek-R1 flagship, refined through massive Reinforcement Learning (RL). Rivaling OpenAI o1/o3 in benchmarks like AIME and Codeforces, it features an "autonomous reflection" mechanism to self-correct reasoning errors. It is the premier choice for building R&D automation, scientific computing, and high-complexity AI agents.
- Context window: 128,000 tokens
- Max output: 128,000 tokens
- Input: text
- Output: text
- Reasoning: Supported
- Tool calling: Supported
- File input: Supported
- Released: 2025-01-20
- Knowledge cutoff: 2024-06-30
Frequently Asked Questions
What is the context window of deepseek-reasoner?
deepseek-reasoner supports a context window of up to 128,000 tokens.
Does deepseek-reasoner support function calling?
Yes. deepseek-reasoner supports tool / function calling.
Does deepseek-reasoner support reasoning?
Yes. deepseek-reasoner is a reasoning-capable model.
What is the knowledge cutoff of deepseek-reasoner?
The knowledge cutoff of deepseek-reasoner is 2024-06-30.