DeepSeek: deepseek-v3.1-terminus API
A flagship Mixture-of-Experts (MoE) model released by DeepSeek on September 22, 2025. As the definitive update to the V3.1 series, "Terminus" (Latin for "end" or "boundary") focuses on refining multilingual consistency and eliminating character glitches reported in earlier builds. It represents the pinnacle of the Hybrid Reasoning Architecture, allowing users to toggle between "Thinking" and "Non-Thinking" modes within a single model. With significant post-training optimizations, it delivers the most stable and robust performance for Code and Search Agents within the V3 family, serving as a reliable production-grade foundation in early 2026.
- Context window: 164,000 tokens
- Max output: 164,000 tokens
- Input: text
- Output: text
- Reasoning: Supported
- Tool calling: Supported
- Released: 2025-09-29
- Knowledge cutoff: 2025-05-31
Frequently Asked Questions
What is the context window of deepseek-v3.1-terminus?
deepseek-v3.1-terminus supports a context window of up to 164,000 tokens.
Does deepseek-v3.1-terminus support function calling?
Yes. deepseek-v3.1-terminus supports tool / function calling.
Does deepseek-v3.1-terminus support reasoning?
Yes. deepseek-v3.1-terminus is a reasoning-capable model.
What is the knowledge cutoff of deepseek-v3.1-terminus?
The knowledge cutoff of deepseek-v3.1-terminus is 2025-05-31.