DeepSeek: DeepSeek V3.1 Terminus

DeepSeek V3.1 Terminus is a text model from DeepSeek, released by the provider on 22 September 2025. It is an update to DeepSeek V3.1 that keeps the original model's capabilities while addressing issues reported by users, including language consistency and agent capabilities. Its knowledge cutoff is 31 March 2025. On api14 it sits in the text category and is identified by the model id deepseek/deepseek-v3.1-terminus.

The model takes text as input and returns text as output. It has a context window of 163,840 tokens and a maximum output of 147,456 tokens, so long inputs and long generations are both possible within those bounds. It supports reasoning, which means you can request reasoning output and control whether reasoning is included in the response. It also supports structured outputs and a response format parameter, so responses can be constrained to a schema rather than free text. Tool calling is available through the tools and tool_choice parameters, which lets the model decide when to call a function or forces a particular tool.

Those capabilities point to a few practical uses. The large context window suits tasks that need to hold a lot of material at once, such as working through long documents, codebases or transcripts in a single request. Reasoning support fits problems where the model benefits from working through steps before answering, for example multi step analysis or debugging. Structured outputs and response format are useful when the result has to be parsed by another program, such as extracting fields into JSON. Tool calling makes the model a reasonable choice for agent style workflows where it needs to call external functions, and the update specifically targets agent capabilities and language consistency, which matters for assistants that must stay in one language across a conversation.

You call it through api14's OpenAI compatible API using the model id deepseek/deepseek-v3.1-terminus. The request parameters it supports are frequency_penalty, include_reasoning, max_tokens, reasoning, response_format, stop, structured_outputs, temperature, tool_choice, tools, top_k and top_p. If you already use an OpenAI style client, you can point it at api14 and set the model field to this id, then pass any of those parameters as needed.

Keep the limits in mind when choosing. Input and output are text only, so if your task involves images or audio, this model is not the right fit. The knowledge cutoff of 31 March 2025 means anything after that date is outside its training data, so for very recent information you would need to supply it in the prompt or use tools to fetch it. The context window and maximum output are large but finite, so extremely long jobs still need chunking.