inclusionAI: Ling 3.0 Flash VL

Ling 3.0 Flash VL is a model from inclusionAI. It builds on Ling 3.0 Flash, a 124B total parameter mixture-of-experts model with 5.5B active parameters, and extends it with native visual perception and advanced visual understanding while further strengthening its language capabilities. The model is listed under the text category, so despite the vision features it is positioned primarily as a language model that can also read images and video.

On the input side, Ling 3.0 Flash VL accepts text, images and video, and it produces text output only. The context window is 262,144 tokens, which is large enough to hold long documents, lengthy conversation histories or extended video-derived context in a single request. The maximum output is 32,768 tokens, so responses can be substantial rather than being cut short at a few hundred tokens. Because the model combines a large context with visual input, it can work over material that mixes prose and visual content without needing a separate pipeline for each.

Practical uses follow directly from those capabilities. A developer could send screenshots, diagrams, charts or scanned pages alongside a question and get a text answer, which suits document question answering, support tooling that needs to read what a user pasted, and extraction tasks where the source is an image rather than clean text. Video input opens up summarisation or querying of recorded material, such as reviewing a screen recording or a clip and asking for a description or a set of action items. The long context supports multi-document analysis, long chat sessions and code or log review where the whole file needs to be in view. Tool calling and structured outputs make it usable as the reasoning step inside an agent or a data extraction workflow.

You call the model through api14's OpenAI-compatible API using the model id "inclusionai/ling-3.0-flash-vl". The supported request parameters are frequency_penalty, include_reasoning, logit_bias, logprobs, max_tokens, min_p, presence_penalty, reasoning, repetition_penalty, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_k, top_logprobs and top_p. That set covers sampling control, deterministic runs via seed, token-level inspection through logprobs and top_logprobs, structured output formats, and tool use with tool_choice and tools. Reasoning can be requested and included in the response through the reasoning and include_reasoning parameters.

The main thing to keep in mind is the output modality: the model reads images and video but returns text only, so it will not generate or edit media. If your task is purely text and does not benefit from visual input, the base Ling 3.0 Flash line may be a closer fit, since the VL variant's distinguishing feature is visual perception. For workloads that need image or video generation, or audio input, you would need a different model.