Gemini 2.5 Flash (preview)

2025-04-17 · Publisher : Google · Domain : Artificial Intelligence

Google launches Gemini 2.5 Flash in preview, its first hybrid reasoning model: reasoning can be turned on or off, and a thinking budget makes it possible to trade off quality, cost and latency.

What changes

First fully hybrid Gemini model: reasoning can be enabled or not depending on the request.
Thinking budget adjustable from 0 to 24,576 tokens (API parameter or slider in AI Studio); at 0, cost and latency remain close to 2.0 Flash with better performance.
Major improvement in reasoning compared with Gemini 2.0 Flash, while keeping speed and cost.
Positioned by Google as the best price-performance ratio in its lineup.

### Specifications - Availability: Gemini API via Google AI Studio and Vertex AI. - Identifier: `gemini-2.5-flash-preview-04-17` (discontinued on July 15, 2025 after the move to the stable version). - The 1 million token context and the 65,536 token output are those of the current stable version (API documentation). ### Pricing The preview pricing appears in an image in the post, which was not extracted. The pricing grid was simplified when the stable version was released (June 17, 2025): the difference between prices with and without thinking was removed, see the "Gemini 2.5 Flash (general availability)" entry. ### Benchmarks No numerical score in the text of the post (comparisons are in charts). ### Key takeaways The introduction of the thinking budget becomes the norm across the lineup: it makes it possible to finely control the cost of agents and batch processing.

Links

Tags : LLM, Google, Gemini, Flash, Raisonnement