inclusionAI: Ling 3.1 Flash

Ling 3.1 Flash is a text model from inclusionAI. It is a hybrid reasoning mixture-of-experts model, with 25B active parameters out of 560B total. The provider released it on 2026-10-02. It sits in the text category, so it is aimed at language tasks rather than image, audio or video work.

The model takes text as input and returns text as output. Its context window is 262144 tokens, and the maximum output is 235929 tokens, so both long inputs and long generations are possible within those bounds. Because it is a hybrid reasoning model, it can produce reasoning output alongside its final answer, and that behaviour can be controlled through the reasoning and include_reasoning parameters. The mixture-of-experts design means only a fraction of the total parameters are active per token, which is relevant if you care about throughput relative to a dense model of similar total size.

Practical use cases follow from the long context and the reasoning capability. Long document analysis, summarisation of large files, and question answering over lengthy source material fit within the 262144 token window. Code and technical work benefit from the reasoning mode, where the model can work through a problem before answering. Tool calling is supported, so the model can be used as the decision layer in agent style workflows that need to call external functions. Structured extraction and classification over large batches of text are also reasonable fits, and the seed parameter allows reproducible sampling when you need consistent outputs across runs.

You call the model through api14's OpenAI-compatible API using the model id "inclusionai/ling-3.1-flash". The supported request parameters are frequency_penalty, include_reasoning, logprobs, max_tokens, presence_penalty, reasoning, repetition_penalty, seed, stop, temperature, tool_choice, tools, top_k, top_logprobs and top_p. If you want token level detail on the output, logprobs and top_logprobs are available. If you want the model to use external functions, pass tools and tool_choice. Sampling behaviour is controlled with temperature, top_p, top_k, repetition_penalty, presence_penalty and frequency_penalty, and stop sequences are supported.

One thing to keep in mind is that the model is text only. If your application needs to process images or audio directly, this model will not accept those inputs. The reasoning mode is optional and can be turned on or off, so you can decide per request whether the extra reasoning output is useful or just adds tokens you do not need. For straightforward short text tasks, a smaller or non-reasoning model may be simpler to operate, while Ling 3.1 Flash is the better fit when context length, reasoning or tool calling are the deciding factors.