for(geeks)
AI

• 2 min read

Google Drops Gemini Sampling Controls in New Flash Models

Gemini 3.6 Flash and 3.5 Flash-Lite are now GA, and Google says temperature, top_p, and top_k are deprecated and ignored.

Image: Hacker News

Google has made Gemini 3.6 Flash and Gemini 3.5 Flash-Lite generally available, and the bigger immediate change for developers may be in the API: temperature, top_p, and top_k are now deprecated and ignored for these models, and Google says future model generations will return an HTTP 400 error if they are supplied.

The same update also blocks prefilled model turns. If the last non-empty turn in a request is a model turn, the API now returns a 400 error. Google recommends moving those controls into system_instruction or using Structured outputs instead.

The two new production models are positioned differently:

  • Gemini 3.6 Flash (gemini-3.6-flash) is aimed at code generation, multimodal reasoning, and multi-step agentic workflows.
  • Gemini 3.5 Flash-Lite (gemini-3.5-flash-lite) is the fastest, lowest-cost model in the 3.5 family, targeted at high-volume parsing, extraction, and structured JSON work.

Google prices Gemini 3.6 Flash at $1.50 per 1M input tokens and $7.50 per 1M output tokens. Gemini 3.5 Flash-Lite comes in at $0.30 per 1M input tokens and $2.50 per 1M output tokens. Both support a 1M-token context window, 64k max output tokens, thinking, and built-in tools including Computer Use.

According to Google, Gemini 3.6 Flash improves multi-step workflow efficiency, code generation, instruction following, and multimodal spatial reasoning, while also cutting output token pricing from $9.00/1M on Gemini 3.5 Flash to $7.50/1M. Gemini 3.5 Flash-Lite is pitched as a stronger migration target from Gemini 2.5 Flash, with benchmark gains including HLE: 18.0% vs. 11.0% and CharXIV: 74.5% vs. 63.7%.

Recommended reading

Kimi K3 Nearly Matches Fable at a Fraction of the Cost

Gemini API migration changes

For apps moving to the new models, Google says developers should:

  • switch model IDs to gemini-3.6-flash or gemini-3.5-flash-lite
  • remove temperature, top_p, and top_k
  • replace thinking_budget with thinking_level such as medium or high
  • remove candidate_count, which is unsupported in Gemini 3.x
  • remove prefilled model turns and rely on server-side conversation state such as previous_interaction_id

Google also says Gemini 3.6 Flash is now the default model behind the Antigravity agent in Gemini Managed Agents. The documentation page was last updated on 2026-07-21 UTC.

Ava Chen

AI Editor

Ava covers the rapidly evolving world of artificial intelligence, from foundational models and research labs to the real-world economics of intelligence. With a background in computational linguistics, she cuts through the hype to find out what actually works. She firmly believes that benchmarks are just marketing until reproduced in the wild.

via Hacker News

// Keep reading