Start building with Gemini 2.5 Flash

0 3 minutes read

Today we are introducing an early version of Gemini 2.5 flash in Inspection Through API Gemini via Google Ai Studio and Vertex Ai. Depending on the famous basis of 2.0 Flash, this new version offers a great upgrade in the possibilities of thinking, while still determines the speed and cost. Gemini 2.5 Flash is the first full hybrid thinking model, giving developers the ability to run or turn off thinking. The model also allows developers to set thinking budgets to find the correct comparison between quality, cost and cumin. Even with Thinking, Developers can maintain rapid speeds of 2.0 flash, and improve performance.

Our Gemini 2.5 models are thinking models, able to think through their ideas before responding. Instead of generating the output immediately, the model can perform a “thinking” process to better understand the claim, break up complex tasks, and plan to respond. In complex tasks that require multiple steps of thinking (such as solving mathematics problems or analyzing research questions), the thinking process allows the model to reach more accurate and comprehensive answers. In fact, Gemini 2.5 Flash strongly leads to solid claims in LMARNA, only second to 2.5 Pro.

2.5 Flash has measures similar to the other leading models of a small part of cost and size.

The most cost -cost thinking model

2.5 Flash continues to drive as a model with the best rate of performance.

Gemini 2.5 Comparison of flash to performance

Gueini 2.5 Flash adds another model to the Google border from the cost to quality.*

Continue granules controls for thinking management

We know that different cases of use have different differentials in quality, cost and cumin. To give developers flexibility, we enabled the preparation Thinking budget This provides precise control of the maximum number of symbols that the model can generate during thinking. The higher budget allows the model to re -improve quality. More importantly, although the budget determines a ceiling on the amount of what can be thought about 2.5 flash, the model does not use the full budget if the claim does not require that.

Improvements in the quality of thinking while increasing the thinking budget.

The model is trained to know the time it takes to think about a specific wave, and thus automatically decides the amount that must be thought about based on the complexity of the perceived task.

If you want to keep the lowest cost and writer while continuing to improve performance more than 2.0 flash, Determine the thinking budget to 0. You can also choose Determine a specific symbolic budget For the thinking stage using a teacher in API or the scrolling bar in Google Ai Studio and in Vertex AI. The budget can range from 0 to 24576 a symbol of 2.5 flash.

The following claims show how much thinking can be used in the default mode of 2.5 Flash.

Claims require low thinking:

Example 1: “Thank you” in Spanish

Example 2: How many provinces do Canada?

Claims require medium thinking:

Example 1: You are a roll of two dice. What is the possibility of adding up to 7?

Example 2: My gym enjoys small hours of basketball between 9-3 pm at MWF and 2-8 pm on Tuesday and Saturday. If you work from 9 to 6 pm 5 days a week and want to play 5 hours of basketball on weekly, create a schedule for me to make everything working.

Claims require high thinking:

Example 1: The beam of the length of L = 3M has a rectangular cross section (width B = 0.1M, height H = 0.2M) is made of steel (E = 200 GPA). It is undergoing a uniform distributor W = 5 kilograms/m completely length and carrying P = 10 kn at its free end. Calculate the maximum bending pressure (σ_max).

Example 2: Write a job evaluate_cells(cells: Dict[str, str]) -> Dict[str, float] That calculates the values of the cells ’cells.

Each cell contains:

Or a formula like "=A1 + B1 * 2" Use +and -and *and/ And other cells.

requirements:

Solve the conversion between cells.

The precede of the operator (*/ before +-).

Discover the courses and raise ValueError("Cycle detected at ").

no eval(). Use only integrated libraries.

Start construction with Gemini 2.5 Flash today

Gueini 2.5 Flash is now available with the possibilities of thinking about the gemini API in Google Ai Studio and Vertex AI, and in a customized drop in the Gemini app. We encourage you to experience thinking_budget Teacher and explore how controlled thinking can help you solve the most 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)