Qwen: Qwen3.5-35B-A3B

Qwen3.5-35B-A3B is a text generation model from Qwen, part of the Qwen3.5 Series. It is a native vision-language model built on a hybrid architecture that combines linear attention mechanisms with a sparse mixture-of-experts design, which the provider describes as achieving higher inference efficiency. The model was released by the provider on 2026-02-25 and is listed on api14 under the text category.

The model accepts text, image and video inputs and returns text output. Its context window is 262144 tokens, and the maximum output is 32768 tokens. Because it handles vision inputs alongside text, it can work with material that mixes prose and visual content rather than text alone. Reasoning behaviour is exposed through the supported parameters, so you can control whether reasoning is included in the response.

The combination of a large context window and image plus video input points to use cases such as summarising or extracting information from long documents that include figures, answering questions about screenshots or diagrams, and processing video content into text descriptions or structured notes. The long context also suits tasks where you need to keep a large body of reference material in the prompt, such as multi-document analysis or extended conversations that must retain earlier detail. Text-only generation tasks, including structured extraction and tool-driven workflows, follow from the same capabilities.

You call the model through api14's OpenAI-compatible API using the model id "qwen/qwen3.5-35b-a3b". The request parameters it supports 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. Tool calling and structured outputs are available through the tools, tool_choice, response_format and structured_outputs parameters, and sampling can be tuned with temperature, top_p, top_k, min_p and the penalty parameters. Log probability data is available through logprobs and top_logprobs, and seed allows you to request reproducible sampling.

The model outputs text only, so tasks that require generating images, audio or video are outside its scope. Its maximum output of 32768 tokens is a constraint to keep in mind when a single response needs to be very long, even though the input context is much larger. If your workload is purely text and does not benefit from vision input or the long context, a simpler text model may be a better fit.