Gemini last models: temperature, top_p, and top_k are deprecated and ignored (ai.google.dev)

Google has announced the general availability of its new AI models, Gemini 3.6 Flash and Gemini 3.5 Flash-Lite, which bring significant advancements to the AI/ML landscape. Gemini 3.6 Flash enhances performance in agentic and multimodal tasks while reducing token usage and costs, compared to its predecessor. Meanwhile, Gemini 3.5 Flash-Lite emerges as the fastest and most cost-effective option in the 3.5 family, ideal for high-throughput execution tasks. Both models support a 1M token context window and the full suite of built-in tools, making them highly versatile for developers. A key aspect of this update includes the deprecation of certain sampling parameters like temperature, top_p, and top_k, which will no longer be recognized by the API. Developers are advised to revise their code accordingly, eliminating these parameters to avoid errors in future model generations. Notable improvements in Gemini 3.6 Flash consist of better token and turn efficiency, enhanced code generation quality, and more reliable instruction adherence, which collectively streamline complex workflows. In summary, these models represent a robust evolution in AI capabilities, promising better performance and cost efficiency for developers in the AI community.