Today we are introducing an early version Flash Gemini 2.5 IN announcement by API Gemini via Google to learn AND Vertex AI. Based on the popular 2.0 Flash foundation, this new version ensures a significant update of the possibility of reasoning, at the same time a priority of speed and costs. Gemini 2.5 Flash is our first fully hybrid reasoning model, which gives programmers the opportunity to turn on or disable their thinking. The model also allows programmers to establish thinking budgets to find the right compromise between quality, costs and delay. Even with Thinking, Developers can maintain a fast speed of 2.0 flash and improve performance.
Our models Gemini 2.5 are thought by models capable of reasoning their thoughts before answering. Instead of generating an output immediately, the model can perform the “thinking” process to better understand the prompt, break down complex tasks and plan your answer. In the case of complex tasks that require many stages of reasoning (such as solving mathematical problems or analysis of research questions), the thinking process allows the model to achieve more accurate and comprehensive answers. In fact, Gemini 2.5 Flash works strongly Hard hints in LmarenaSecond only up to 2.5 Pro.
2.5 Flash has comparable indicators with other leading models for a fraction of costs and size.
Our most profitable thinking model
2.5 Flash continues as a model with the best price -performance ratio.
Gemini 2.5 Flash adds another model to Pareto Frontier Pareto Frontier to quality.*
Fine -grained controls for thinking management
We know that various cases of use have different compromises of quality, costs and delays. To provide programmers with flexibility, we enabled the setting Thinking budget This offers a fine -grained control over the maximum number of tokens that the model can generate during thinking. The higher budget allows the model to justify further improvement in quality. Importantly, the budget sets a limit of how much 2.5 flash can think, but the model does not use the full budget if Monik does not require it.
Improving the quality of reasoning as the thinking budget increases.
The model is trained to know how to think about a given poem, and therefore automatically decides how to think on the basis of the perceived complexity of tasks.
If you want to maintain the lowest costs and delay, it also improves performance above 2.0 flash, Set the thinking budget to 0. You can also choose Set a specific token budget For the phase of thinking using the API parameter or a slider at Google Ai Studio and Vertex AI. The budget can range from 0 to 24576 tokens for 2.5 flash.
The following hints show how much reasoning can be used in the default mode of 2.5 flash.
Prompts requiring low reasoning:
Example 1: “Thank you” in Spanish
Example 2: How many provinces does Canada have?
MONITS requiring average reasoning:
Example 1: You throw two bones. What are the probability that they add up to 7?
Example 2: My gym has basketball hours from 9-13 in MWF and from 14 to 20 on Tuesday and Saturday. If I work 9-16, 5 days a week and want to play 5 hours in basketball on weekdays, create a schedule for me to make everything work.
Prompts requiring high reasoning:
Example 1: A bracket beam with a length l = 3m has a rectangular cross -section (width b = 0.1 m, height h = 0.2 m) and is made of steel (E = 200 GPA). It is subjected to an evenly distributed load in = 5 kN/m along the entire length and point load p = 10 k at the free end. Calculate the maximum bending stress (σ_max).
Example 2: Write a function evaluate_cells(cells: Dict(str, str)) -> Dict(str, float) This calculates the values of the spreadsheet cells.
Each cell contains:
- Or such a formula
"=A1 + B1 * 2"using+IN-IN*IN/and other cells.
Requirements:
- Solve the relationships between cells.
- The first attitude of the handle operator (
*/before+-).
- Detect cycles and raise
ValueError("Cycle detected at.") |
- NO
eval(). Use only built -in libraries.
Start construction with Bliglars 2.5 Flash today
Gemini 2.5 Flash with thinking possibilities is now available in view through API Gemini IN Google to learn and in Vertex AIand in dedicated development in Gemini application. We encourage you to experiment with thinking_budget Parameter and examine how controlled reasoning can help solve more complex problems.
from google import genai
client = genai.Client(api_key="GEMINI_API_KEY")
response = client.models.generate_content(
model="gemini-2.5-flash-preview-04-17",
contents="You roll two dice. What’s the probability they add up to 7?",
config=genai.types.GenerateContentConfig(
thinking_config=genai.types.ThinkingConfig(
thinking_budget=1024
)
)
)
print(response.text)
Find detailed references to the API interface and thinking guides in ours Programmers' documents or start with Code examples With Gemini cookbook.
We will continue to improve Flash Gemini 2.5, and more soon before we generally make it available for full use of production.
*Model prices come from the company's artificial analysis and documentation
