Gemini 2.5 Pro Exp 03 25: Preview Release Details
Google released Gemini 2.5 Pro Exp 03 25 on March 25, 2025. This experimental model features text and image inputs with thinking mode enabled by default.
Release Overview and Modalities
The original announcement date for the release of gemini-2.5-pro-exp-03-25 was March 25, 2025. This article is published later than that historical event date. The model is identified as a public experimental version within the Gemini family provided by Google. It accepts both text and image data as input modalities. The output modality is restricted to text. The source does not specify a context window size or maximum output token limit for this specific experimental build.
Thinking Mode Configuration
A defining characteristic of this preview release is that thinking mode is always on by default. The official changelog describes the model as having this feature enabled without requiring user configuration to activate it. The source does not provide technical details on how the thinking process operates internally or what specific computational steps occur during this phase. It also does not state whether this mode can be disabled by the user in this specific experimental version. The documentation notes that open weights are not confirmed for this model, which records the absence of verified open-weight status rather than establishing local installation requirements or API exclusivity.
Evaluation Procedure and Limitations
Engineers evaluating this model should focus on its documented capabilities. A practical evaluation procedure involves testing the model with mixed text and image inputs to measure response consistency. Users should compare the output quality against their specific application requirements. Since the model is experimental, performance may vary. The source confirms the release date and basic modalities but does not provide benchmark scores, pricing information, or performance rankings relative to other models. It remains unknown how the default thinking mode affects latency or cost compared to non-thinking modes, as no such data is present in the source facts.
- Test input handling with both text and image files.
- Measure response time to observe the impact of the default thinking mode.
- Validate output accuracy against known ground truth datasets.
- Compare results with previous stable versions if available in your environment.
Primary sources
Frequently asked questions
Is the thinking mode optional in Gemini 2.5 Pro Exp 03 25?
No, the source states that thinking mode is always on by default for this experimental model. It does not indicate that users can toggle this feature off.
Can I install Gemini 2.5 Pro Exp 03 25 locally?
The source indicates that open weights are not confirmed for this model. It does not explicitly state whether local installation is impossible or mandatory API usage is required, only that open weights are unverified.