Qwen: Qwen3 235B A22B Thinking 2507

Qwen3 235B A22B Thinking 2507 is a text model from Qwen, released by the provider on 2025-07-25. It is an open-weight Mixture-of-Experts (MoE) language model built for complex reasoning tasks, and it activates 22B of its 235B parameters per forward pass. The "Thinking" label and the reasoning parameters it exposes indicate that the model is designed to work through problems step by step rather than answer in a single pass.

The model accepts text input and returns text output, so it fits text-only pipelines. It has a context window of 128,000 tokens and a maximum output of 16,384 tokens, with a knowledge cutoff of 2025-06-30. Because it is a reasoning-oriented model, responses may include a reasoning phase, and the include_reasoning and reasoning parameters let you control how that is handled in the response you receive.

Those capabilities point to use cases where careful multi-step thinking matters: mathematical and logical problem solving, code analysis and generation, structured extraction from long documents, and multi-step planning or decision support. The 128,000 token context is large enough to hold substantial source material alongside instructions, which suits tasks such as reviewing long specifications, summarizing or querying lengthy reports, and working through problems that require the model to keep many constraints in view at once. Tool calling support also makes it usable in agent-style workflows where the model decides when to call a function.

You call the model through api14's OpenAI-compatible API using the model id "qwen/qwen3-235b-a22b-thinking-2507". The supported request parameters are frequency_penalty, include_reasoning, max_tokens, presence_penalty, reasoning, stop, temperature, tool_choice, tools, top_k and top_p. Since the interface is OpenAI-compatible, you can point an existing OpenAI client at api14 and change the model id, then tune sampling with temperature, top_p or top_k, cap the response with max_tokens, and pass tools and tool_choice when you want function calling.

Keep the output limit in mind: max_tokens caps at 16,384, so tasks that need a very long single response may require splitting the work across calls. The model is text-only, so if your input includes images or audio you would need a different model. For straightforward, high-volume text tasks where deep reasoning is not needed, a lighter model may be a better fit, while this one is aimed at problems that benefit from explicit reasoning.