Gemini 3.6 Flash Launches as Temperature and Top_P Parameters Are Deprecated
Gemini last models: temperature, top_p, and top_k are deprecated and ignored
I am excited to announce the general availability of Gemini 3.6 Flash and Gemini 3.5 Flash-Lite, offering superior agentic performance and lower costs. These new models introduce critical API changes, specifically the deprecation of temperature, top_p, and top_k parameters, which are now ignored. Developers must migrate to the Interactions API and remove prefilled model turns to ensure their applications function correctly with these advanced capabilities.
Starting with Gemini 3.6 Flash and Gemini 3.5 Flash-Lite, the sampling parameters temperature, top_p, and top_k are deprecated and ignored, returning an error in future model generations.
- salamo
Possible reasons:
- They might be dynamically adjusting these at inference time [1]. For example, start with a low temperature and generate samples with increasingly high temperatures until one of them passes some quality gate.
- They don't want you to fine-tune on high temperature completions (rejection fine-tuning). You could call this "rejection fine-tuning rejection".
[1] https://rlhfbook.com/c/09-rejection-sampling#related-best-of...
- franky47
> To improve determinism, define a system instruction with explicit rules for your specific use case.
"Please be deterministic".
- aesthesia
My guess is that RL training being done with particular generation parameters makes models much more brittle to changes in these parameters, and that's why we're seeing changes like this across model providers. But I don't really know.
- bob1029
> To improve determinism, define a system instruction with explicit rules for your specific use case.
What if I want to do the other thing? When performing research with many sub agents, having a lot of diversity in the hypotheses is a big deal. If my 5 parallel sub agents all produce the same conclusion I might as well have only ran one.
The latest OAI models have done the same thing. I'm currently adding random variation to prompts to compensate for the lack of higher temperature sampling.
- marcinjachymiak
It's not surprising since these are all reasoning models now. Token sampling/decoding isn't exposed because the reasoning models are trained around specific configurations. That's why it's been replaced with "reasoning effort" instead (low, medium, high "thinking" levels).