O Google just launched o Gemini 2.5 Flash - an artificial intelligence (AI) hybrid reasoning in preview that matches the o4-mini, surpasses the Claude 3.5 Sonnet in thinking/STEM benchmarks and introduces a new “thinking budget” to optimize cost versus quality.
ADVERTISING
details of Gemini 2.5 Flash
- 2.5 Flash demonstrates significant increases in reasoning over its predecessor (2.0 Flash), with a controllable thought process to turn the feature on or off.
- The model performs strongly in reasoning, STEM and visual reasoning benchmarks, despite costing much less than its rivals.
- Developers can also set a “thinking budget” (up to 24 tokens), which adjusts the balance between response quality, cost, and speed.
- It is available via API through the Google AI Studio and Vertex AI, and is also appearing as an experimental option within the app Gemini.
Why is it important
A OpenAI may have dominated the conversation this week, but the Google is rolling out new features alongside them. Controllable and budgetable thinking is an interesting customization, with users being able to trigger the feature only when a task requires it – unlocking accessible, high-volume use cases and reserving the “thinking” for more complex work.
Read also
