Gemini 3 Flash Preview Thinking API

Input / 1M tokens$0.44~~$0.50~~

Output / 1M tokens$2.64~~$3.00~~

LLMGoogle12% below officialReasoning

Loading playground…

About Gemini 3 Flash Preview Thinking

Gemini 3 Flash Preview Thinking by Google is available on API Many from $0.44 per million input tokens and $2.64 per million output tokens, and you call it with the same API key you use for every other model.

It accepts OpenAI-style chat requests with streaming and tools.

You pay only for successful results: failed or rejected requests are refunded automatically, and every call is recorded in your console with its exact cost.

Related models

New−12%

Gemini 4 Argongemini-4-argon

LLMGoogle128K contextPrompt caching

$1.76 in · $8.80 outTry it

−12%

Gemini 2.5 Flashgemini-2.5-flash

LLMGoogle

$0.264 in · $2.20 outTry it

−12%

Gemini 2.5 Flash Litegemini-2.5-flash-lite

LLMGoogle

$0.088 in · $0.352 outTry it

−12%

Gemini 2.5 Flash Nothinkinggemini-2.5-flash-nothinking

LLMGoogleReasoning

$0.264 in · $2.20 outTry it

−12%

Gemini 2.5 Flash Thinkinggemini-2.5-flash-thinking

LLMGoogleReasoning

$0.264 in · $2.20 outTry it

−12%

Gemini 2.5 Progemini-2.5-pro

LLMGoogle

$1.10 in · $8.80 outTry it

−12%

Gemini 2.5 Pro Nothinkinggemini-2.5-pro-nothinking

LLMGoogleReasoning

$1.10 in · $8.80 outTry it

−12%

Gemini 2.5 Pro Thinkinggemini-2.5-pro-thinking

LLMGoogleReasoning

$1.10 in · $8.80 outTry it

Frequently asked questions

How much does the Gemini 3 Flash Preview Thinking API cost?

Gemini 3 Flash Preview Thinking starts at $0.44 per million input tokens and $2.64 per million output tokens on API Many. The price table on this page lists every tier; you are charged only for successful requests.

How do I call Gemini 3 Flash Preview Thinking?

Create an API key in the console, then send a request to /v1/chat/completions with "model": "gemini-3-flash-preview-thinking". The API tab has copy-ready examples in cURL, Python and Node.js.

Is this the official Gemini 3 Flash Preview Thinking model?

Yes. Requests are forwarded to the model maker or an authorized channel with the same parameters, so the output matches the official API.

What happens if a request fails?

Failed, rejected or timed-out requests are not charged. Reserved balance is released automatically and the reason is shown in the call record.