Gemini 3.6 Flash is the fast, everyday model in Google DeepMind's Gemini 3 family, built for coding, agents and document-heavy work. Its signature move: finishing the same task with 17% fewer output tokens than its predecessor. Trying it costs nothing in the Gemini app or Google AI Studio, and API output pricing dropped to $7.50 per million tokens. The payoff: near-Pro quality at Flash speed.
Gemini 3.6 Flash uses 17% fewer output tokens than Gemini 3.5 Flash on the Artificial Analysis Index, and up to 65% fewer on the DeepSWE coding benchmark. It also completes multi-step jobs with fewer reasoning steps and tool calls, which counts when an intelligent agent runs from morning to night.
The spec sheet keeps up: a million tokens of context on input, 64K on output, and native multimodal input. You drop in a video or a 300-page PDF, ask your questions, and it answers.
Google treats this model as its production workhorse: on DeepSWE, the score climbs from 37% to 49%, with fewer unwanted code edits and fewer execution loops.
Company feedback points the same way. Figma picked it to speed up prototyping in Figma Make, law firms review M&A paperwork with it, and financial analysts use it to dig proof points out of dense transaction documents. It also became the default model in the Gemini app and in Antigravity for coding.
Timing mattered here: the release filled the wait for Google's next Pro model while pre-training for Gemini 4 was getting underway. To see where those scores sit against OpenAI or Anthropic, the LLM model rankings keep things honest.
Two free routes exist: the Gemini app, where the model answers by default, and the rate-limited free tier of the API in Google AI Studio (free-tier prompts can feed Google's product tuning, worth knowing before you paste a contract in). Production traffic runs on the standard API, with thinking tokens billed as output.
These figures shift with each Google announcement, so check the official pricing page before budgeting anything.
| Model | Role | API price per million tokens |
|---|---|---|
| Gemini 3.6 Flash | Coding, agents, multimodal work | $1.50 input, $7.50 output |
| Gemini 3.5 Flash-Lite | High volume, low latency | $0.30 input, $2.50 output |
| Gemini 3.5 Flash Cyber | Finding and fixing security flaws | Restricted pilot, not publicly sold |
Yes, in two ways: the Gemini app runs it by default for fast answers, and Google AI Studio includes a free API tier capped by rate limits. Both routes are built for testing and prototyping rather than production traffic; once an app goes live, the paid API takes over.
They count toward the output total and cost the same as regular response tokens, with no separate line on the invoice. Longer reasoning therefore means a bigger bill; context caching, billed separately, helps cut the cost of repetitive prompts.
Only the free API tier carries that caveat: Google may reuse those prompts to improve its products. A paid key or a deployment through Google Cloud's enterprise platform avoids the reuse, which makes it the sensible route for confidential documents and client data.
Verdict: Quieter agents and a lighter API bill: keep a Pro model in reserve for the heaviest refactors, and the rest of the day belongs to development teams, analysts and anyone chaining model calls for hours on end.
