Google: Gemini 2.5 Pro Preview 06-05

Google: Gemini 2.5 Pro Preview 06-05 is a text generation model from Google, offered through api14 under the model id google/gemini-2.5-pro-preview. It is part of the Gemini 2.5 Pro line, which Google describes as its state-of-the-art model built for advanced reasoning, coding, mathematics and scientific tasks. The model uses what Google calls thinking capabilities, meaning it can reason through a response before producing it, which is intended to improve accuracy on problems that need multiple steps.

On the input side, the model accepts text, images, audio and files, so you can send mixed content in a single request. Output is text only. The context window is 1,048,576 tokens, which is large enough to hold long documents, codebases or extended conversations without chunking. The maximum output length is 65,536 tokens, and the knowledge cutoff is 2025-01-31, so anything after that date is outside what the model was trained on.

Those capabilities point to a few practical uses. The long context suits working over entire repositories, lengthy specifications or transcripts in one call. Image and audio input let you build features that reason over screenshots, diagrams or recorded speech and return written results. The reasoning behaviour is a good fit for coding assistance, mathematical problem solving and scientific analysis, where a model benefits from working through intermediate steps. Structured outputs and tool calling also make it usable as the reasoning component inside an automated pipeline rather than only as a chat interface.

Because api14 exposes an OpenAI-compatible API, you call this model the same way you call OpenAI models, setting the model field to google/gemini-2.5-pro-preview. Supported request parameters are include_reasoning, max_tokens, reasoning, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools and top_p. So you can control sampling with temperature and top_p, cap length with max_tokens, constrain output with response_format or structured_outputs, define functions with tools and tool_choice, and use seed and stop for more deterministic or bounded generation. The reasoning and include_reasoning parameters relate to the model's thinking behaviour, letting you request or surface that reasoning.

One thing to keep in mind is that output is text only, so if you need generated images or audio you will need a different model. The knowledge cutoff of 2025-01-31 means questions about more recent events rely on what you supply in the prompt. For simpler, high-volume tasks where the long context and reasoning steps are unnecessary, a smaller or faster model may be a better fit.