Gemini 3.7 Flash is Google's new coding-and-agents model, released August 13, 2026, just three weeks after Gemini 3.6 Flash. It ships at an introductory price of $0.75 per million input tokens and $3.75 per million output tokens — half of 3.6 Flash's rate — and it's already live in Google AI Studio, Android Studio, Google Antigravity, and the Gemini Enterprise platform.
Short answer: Gemini 3.7 Flash launched August 13, 2026, priced at $0.75/M input and $3.75/M output tokens through December 31, 2026 (rising to $1.50/$7.50 after). It has a 1,048,576-token context window, a 65,536-token output cap, and reads text, images, video, audio, and PDFs. Access it in Google AI Studio, Android Studio, or via the API as
gemini-3.7-flash.
In my testing, I pulled the pricing and benchmark figures below straight from Google’s own launch post and cross-checked the token limits and model ID against the current Gemini API pricing and model docs, because launch-day numbers tend to get restated inaccurately by aggregator sites within a day or two. Here's what 3.7 Flash actually costs, how it stacks up against Google's other current Flash models, and the exact steps to start using it.
What you'll need
For casual testing, a free Google account gets you into AI Studio, where the model is selectable from the dropdown with no card on file. If you're building against the API, you need a Google AI Studio or Vertex AI project with billing enabled, since Flash is pay-per-token from the first call — there's no separate free tier for this specific model beyond AI Studio's rate-limited playground. Consumers who just want to chat with it need a Google AI Pro or Ultra subscription, since it currently reaches the public only through Spark, Google's agent inside Chrome, and Spark isn't available in the EEA, UK, Switzerland, or Nigeria yet. If you're doing agentic coding work, it also helps to already have Android Studio or Google Antigravity installed, since both wire the model in directly.
Step-by-step: Gemini 3.7 Flash
1. Pick the surface that matches your task
Google AI Studio for quick prompting and prototyping, Android Studio for mobile-app coding, Google Antigravity for broader agentic development, the Gemini Enterprise Agent Platform or app for business workflows, or the raw API for anything custom. All of these had Flash live at launch, so the choice is about the job, not availability.
2. Confirm the model picker shows 3.7, not 3.6
Open the model dropdown in AI Studio and look for "3.7 Flash" specifically. Google kept 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber available at the same time, and they sit in the same list — it's easy to keep running an older Flash release without noticing the version number.
3. Generate an API key and call the model by its ID
Sign in at Google AI Studio, generate a key, and call the model as gemini-3.7-flash. A minimal Python call looks like this:
“python from google import genai client = genai.Client(api_key="YOUR_KEY") resp = client.models.generate_content( model="gemini-3.7-flash", contents="Summarize this in three bullet points: ..." ) print(resp.text) “
4. Set a thinking level instead of leaving it on default
It supports low, medium, and high thinking configurations that trade reasoning depth against latency and cost. In my testing, low handled a short rewrite and a basic regex about as well as high did, but noticeably faster; switching to high on a multi-file debugging task produced a more careful fix, at a slower response.
5. Feed it real files, not just text
Since it accepts images, video, audio, and PDFs alongside text, and returns up to 65,536 output tokens, it's worth testing on the actual documents or screenshots you deal with — a long PDF or a screen recording — rather than only plain prompts, since that's where the 1M-token context window earns its keep.
6. Watch the pricing clock
The $0.75/$3.75 per-million-token rate is introductory and expires December 31, 2026. From January 1, 2027 it doubles to $1.50 input and $7.50 output. If you're budgeting a project that runs into next year, plan around the higher number, not the launch-week one.
Example prompts you can copy
These lean on what Google specifically built 3.7 Flash for — coding and agentic workflows:
- "Here's my repo structure [paste or attach]. Find the bug causing [symptom], fix only that file, and explain your reasoning before the diff." (Tests debugging accuracy, where Google's own FrontierCode 1.1 numbers show the biggest jump over 3.6 Flash.)
- "Plan and execute a multi-step task: open this webpage, extract the pricing table, and return it as JSON." (Tests agentic tool use and multi-step planning.)
- "Review this pull request for correctness bugs only, not style, and rate your confidence on each finding." (Try it at low vs. high thinking and compare the diffs.)
- "Read this 40-page PDF and list the three numbers I should double-check before repeating them externally." (Tests the 1M-token context window on a real document.)
- "Watch this screen recording and describe, step by step, what the user did wrong." (Tests video input, one of 3.7 Flash's supported modalities.)
Common mistakes to avoid
The first mistake I'd flag: assuming "Flash" in the model picker means 3.7. Google left 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber selectable in the same dropdown, so check the version number, not just the family name. Second, treating this as a brand-new model from scratch — Google's own post describes 3.7 Flash as algorithmic improvements to 3.6 Flash's reasoning, not a new pretraining run, so don't expect it to feel like a different generation on tasks 3.6 Flash already handled well. Third, running everything at high thinking by default when a task doesn't need it; in my testing, low thinking was plenty for routine rewrites and cut wait time noticeably. Fourth, budgeting next year's costs off the launch-week rate — the $0.75/$3.75 pricing is introductory and doubles on January 1, 2027. Fifth, expecting consumer access outside of Spark; if you just want to type into a chat box today, AI Studio's playground is the actual front door, not a phone app.
Gemini 3.7 Flash vs. Google's other current Flash models
| Model | Input $/M tokens | Output $/M tokens | Context window | Notable |
|---|---|---|---|---|
| Gemini 3.7 Flash | $0.75 (intro, through Dec 31, 2026) | $3.75 (intro) | 1,048,576 tokens | Coding/agent focus; FrontierCode 1.1 Main 43.6% vs. 3.6 Flash's 34.4% |
| Gemini 3.6 Flash | $1.50 (its own list rate) | $7.50 | 1,048,576 tokens | Previous Flash release, three weeks earlier; still selectable |
| Gemini 3.5 Flash-Lite | Lower than 3.6 Flash (Google's cheapest current tier) | Lower than 3.6 Flash | 1,048,576 tokens | Google's most cost-efficient model, not the strongest coder |
| Gemini 3.5 Flash Cyber | Same family as 3.5 Flash-Lite | Same family | 1,048,576 tokens | Tuned variant with added cyber-defense safeguards |
Pricing confirmed against Google’s launch post and the current Gemini API pricing page on 2026-08-14. On Google's own benchmark set, 3.7 Flash's biggest jump over 3.6 Flash is in coding and agent tasks — DeepSWE v1.1 climbs from 49.0% to 65.3%, and AutomationBench nearly doubles from 17.0% to 30.4% — while costing half as much per token during the introductory window.
Tools that make this easier
If you're deciding whether to switch your default Gemini model or just try 3.7 Flash alongside what you already use, my how to use Gemini guide covers the account setup this article assumes you already have, and how to use Google AI Studio walks through the API-key path in more depth. If you're weighing Gemini against the other major assistants, Gemini vs. ChatGPT, Claude vs. Gemini, and DeepSeek vs. Gemini cover the head-to-head comparisons. I ran the same kind of pricing-and-benchmark check on Anthropic's current flagship in Claude Opus 5: pricing, benchmarks, and setup guide, useful if you're comparing across vendors rather than just within Google's lineup. For coding specifically — 3.7 Flash's strongest category — see how it stacks up in best AI tool for code. My starter kit for AI is the place to start if you haven't set up any model accounts yet, and how we test AI tools explains how I verify numbers like these before publishing them.
Where this leaves you
Gemini 3.7 Flash is a real jump on coding and agent tasks, not just a repricing exercise — Google's own benchmarks show DeepSWE v1.1 up 16 points and AutomationBench nearly doubling over 3.6 Flash, while the introductory rate is half of 3.6 Flash's list price. The thinking-level toggle is the setting I'd actually use day to day: low for routine work saves real wait time, and bumping to high on something that matters is a one-parameter change. If you're already in AI Studio, Android Studio, or Antigravity, it's one dropdown selection away. If you're a consumer without a Pro or Ultra subscription, you'll need to wait for it to reach a wider surface than Spark — worth checking back on before assuming it's out of reach.
Last updated: August 14, 2026 · By Vishal Swami, Founder & Lead AI Reviewer, AISagely
Frequently Asked Questions
Gemini 3.7 Flash: is it free?
You can try it for free in Google AI Studio's rate-limited playground, and consumers with a Google AI Pro or Ultra subscription get it through Spark. The API isn't free — it's pay-per-token at $0.75/M input and $3.75/M output through December 31, 2026, billed to whichever Google Cloud or AI Studio project you attach a card to.
How long does it take to start using it?
Through AI Studio, a couple of minutes — open the model picker and select "3.7 Flash." Through the API, expect 10–15 minutes to create a billing-enabled project, generate a key, and send a first test call with gemini-3.7-flash as the model string.
What is the easiest way to try it?
Open Google AI Studio, select 3.7 Flash from the model picker, and run a real task through it — a coding fix or a long PDF — rather than a generic chat prompt. That's the fastest way to judge whether the upgrade from 3.6 Flash matters for what you actually do.
Is Gemini 3.7 Flash better than Gemini 3.6 Flash?
On coding and agent benchmarks, yes and by a wide margin — DeepSWE v1.1 jumps from 49.0% to 65.3% and AutomationBench nearly doubles, per Google's own numbers. It's built on the same underlying pretraining as 3.6 Flash with algorithmic reasoning improvements layered on, not a ground-up new model, so gains outside coding and agent work are likely smaller.
How does Gemini 3.7 Flash compare to Gemini 3.5 Flash-Lite on price?
Flash-Lite is Google's cheapest current model and stays below 3.7 Flash's rate, but it's not built for the same coding-and-agent workloads. Flash-Lite is the better default for high-volume, simple tasks where cost per call matters most; 3.7 Flash earns its higher price on debugging, multi-step agent work, and long-document reasoning specifically.