How to set the Gemini 2.5 Thinking Budget?

[ath]

For the Anthropic models I have the option to turn thinking on & off, and I also can control the thinking budget. Yet when I try to make this setting with Flash 2.5 (preview) I can’t find it. Any tips? Do I first have to activate this somehow?

[LeeJiaMing1192]

Ah I also got stuck on the same problem for a while but I found out how to control the thinking budget and other parameters like this:

client = genai.Client(api_key="") 


model_name = "gemini-2.5-flash" 

prompt_text = ""

# Example: Setting a thinking budget of 1024 tokens
response = client.models.generate_content(
    model=model_name,
    contents=prompt_text,
    config=types.GenerateContentConfig(
        thinking_config=types.ThinkingConfig(
            thinking_budget=1024  
            
           temperature=0.8, 
        top_k=40,      
        # top_p=0.9,    # You might use top_p instead of or in conjunction with top_k
        max_output_tokens=200
        )
    )
)

print(response.text)

For the thinking budget configuration I had only tried for the latest gemini 2.5 models, for other Gemini models you might wanted to check it out.Also i think reading Gemini documents will be very helpfull : https://ai.google.dev/gemini-api/docs/quickstart. I hope this helps you !

[ath]

This looks like a Python script that makes use of a Google lib to prompt their AIs.
I could easily do this in Haskell or Rust manually, but I had in mind: “Is this available in the LibreChat UI?”

For the Anthropic model it’s pretty easy to do. See the red arrows:

[danny-avila]

For LibreChat, an update was just made that supports this

[ath]

Okay great, will give this a try. Today I’ve updated my installation, about 3h before this new update.