← All models

Gemini 2.5 Pro Experimental

Published Publisher Googlegemini-2.5-pro-exp-03-25

Model Positioning and Release Date

The public experimental version of Gemini 2.5 Pro, released by Google on March 25, 2025, is a multimodal thinking model designed for complex tasks. The official release record states that thinking mode is always enabled by default in this version. (blog.google)

Capabilities and Uses

  • Reasoning and coding: Designed for reasoning in mathematics, science, and complex problems; it can also be used for code generation, transformation, and editing, as well as developing web applications and code agents. (blog.google)
  • Multimodality and long context: At launch, it supported processing information such as text, audio, images, video, and code repositories, with an officially announced context window of 1 million tokens. The 2 million tokens mentioned in the announcement was a future plan at the time and should not be considered a confirmed launch specification for this experimental version. (blog.google)

Limitations and Availability

This is an experimental model, and its output may contain errors; important results should be checked by a human. Google's Gemini API changelog shows that gemini-2.5-pro-exp-03-25 was shut down on June 26, 2025; it should not be treated as the same endpoint as similarly named preview or stable version IDs. (aistudio.google.com)

Pricing

Google's Gemini API changelog states that the experimental version was available on the free tier in April 2025; this is a historical availability condition, not a current service commitment. Separate paid input and output token rates and currency for this version are unverified; prices for other versions are not applied. Date checked: November 19, 2026. (aistudio.google.com)

Official Sources