Gemini 3.6 Flash costs $1.50 per 1M input tokens and $7.50 per 1M output tokens. That’s cheaper than the model it replaces. Google shipped the refresh on July 21, 2026, and for anyone watching an API bill the headline is simple: the workhorse Flash tier got faster and less expensive at the same time.
This guide breaks down every number you need to budget: the per-token rates, what context caching adds, what the free tier covers, and a worked example so you can predict monthly spend before you write a line of code. Every figure here comes from Google’s own Gemini API pricing docs and the launch announcement. If you want the full model overview first, read what is Gemini 3.6 Flash.
One naming note before the numbers. The workhorse jumped to version 3.6, but the cheaper Flash-Lite tier stays at 3.5. Same launch, mixed versioning. Keep that straight and the pricing lines up cleanly.
Gemini 3.6 Flash pricing at a glance
| Item | Cost |
|---|---|
| Input | $1.50 per 1M tokens |
| Output (includes thinking tokens) | $7.50 per 1M tokens |
| Context caching, cached input read | $0.15 per 1M tokens |
| Context caching, storage | $1.00 per 1M tokens per hour |
| Free tier | Yes, via Google AI Studio (rate-limited) |
Two things to flag right away. The output price includes thinking tokens, so the reasoning the model does before it answers is billed at the $7.50 output rate, not on a separate line. And context caching has two parts: a cheap read rate for cached input, plus an hourly storage fee for as long as you keep the cache warm. Caching saves money on repeated context, but it isn’t free, so the savings only show up when you reuse the same prompt prefix a lot.
How it compares to Gemini 3.5 Flash
The output rate dropped from $9.00 per 1M tokens on 3.5 Flash to $7.50 on 3.6 Flash. That alone is about 17% cheaper per output token.
The bigger story is what Google reports on the DeepMind Flash model page: 3.6 Flash produces roughly 17% fewer output tokens for the same task. It reasons in fewer steps and makes fewer tool calls on multi-step work. So you’re paying a lower rate on a smaller pile of tokens. The two levers compound, which we’ll show with real arithmetic below.
If you’re still budgeting against the old model, the Gemini 3.5 Flash pricing breakdown is worth a side-by-side read. For a full feature and benchmark comparison, see Gemini 3.6 Flash vs 3.5 Flash. The short version for your wallet: 3.6 Flash costs less to run than the model it supersedes, and the gap widens on longer, agentic workloads where the token savings add up.
What the free tier gets you
There’s a real free tier, available through Google AI Studio. You can call Gemini 3.6 Flash without a credit card and prototype for free.
Be honest with yourself about the limits. The free tier is rate-limited, so it’s built for prototyping and light testing, not production traffic. And Google may use free-tier data to improve its products. That means the free tier is the wrong place for sensitive, proprietary, or customer data. When you move to anything real, move to a paid key where those data-use terms don’t apply.
For the full walkthrough of what the free tier allows and how to set it up, see how to use Gemini 3.6 Flash for free. There is no “unlimited free” option here, so plan to graduate to paid usage once you ship.
A worked cost example
Say you run an agent that reads a lot and writes a little: 2M input tokens and 500k output tokens per day. That’s a realistic shape for a retrieval-heavy assistant that pulls in documents and returns short answers.
Here’s the daily bill on Gemini 3.6 Flash:
- Input: 2M tokens times $1.50 per 1M = $3.00
- Output: 500k tokens times $7.50 per 1M = $3.75
- Daily total: $6.75
- Monthly (30 days): about $202.50
Now the compounding. Suppose the same task on 3.5 Flash emitted 600k output tokens. On 3.6 Flash the 17% reduction brings that down to roughly 500k, which is what we budgeted above. Compare just the output line:
- 3.5 Flash output: 600k times $9.00 per 1M = $5.40 per day
- 3.6 Flash output: 500k times $7.50 per 1M = $3.75 per day
That’s $1.65 less per day, or about $49.50 a month, on the output side alone. The output bill drops around 31%, even though the price cut and the token reduction are each only about 17%. Cheaper tokens times fewer tokens is where the real savings live.
Flash-Lite is even cheaper
If your task is simple, there’s a tier below the workhorse. Gemini 3.5 Flash-Lite costs $0.30 per 1M input tokens and $2.50 per 1M output tokens, with caching at $0.03 per 1M plus the same $1.00 per 1M per hour storage. It runs fast too, around 350 output tokens per second.
Run the same 2M input and 500k output workload on Flash-Lite and the daily bill is about $1.85, roughly $55.50 a month. That’s around 70% cheaper than 3.6 Flash for the identical token counts.
The catch is quality. Flash-Lite is tuned for high-volume, low-complexity work: classification, extraction, routing, short structured replies. It isn’t a weaker version of the same thing; it’s a different point on the cost, quality, and latency curve. Reach for it when the task is simple and the volume is high, and keep 3.6 Flash for the harder reasoning. For the full picture, read what is Gemini 3.5 Flash-Lite, then Gemini 3.5 Flash-Lite vs 3.6 Flash to pick the right tier per route.
How to control and measure your costs
Four levers keep the bill down:
- Cache repeated context. If every request shares a long system prompt or the same reference documents, context caching charges $0.15 per 1M on cached reads instead of $1.50. Just remember the $1.00 per 1M per hour storage fee, so caching pays off when you reuse the prefix often within the hour, not once.
- Trim prompts. Output is five times the price of input, but bloated input still adds up at scale. Cut boilerplate, and cap max output tokens so a runaway response can’t surprise you.
- Route simple tasks to Flash-Lite. Send classification and extraction to the cheaper tier and reserve 3.6 Flash for multi-step reasoning. Mixed routing often beats picking one model for everything.
- Measure real token usage. Estimates lie. The
usageMetadatain every Gemini response tells you exactly how many input, output, and thinking tokens a call burned, and that’s the number your invoice is built on.
That last point is where an API client earns its keep. In Apidog you can build the POST request to the Gemini API, store your key in an environment variable, and send real calls against your real prompts. Read the usageMetadata fields straight off the response to see actual token counts per request, then add assertions so a prompt change that doubles output tokens fails a test instead of quietly doubling your bill. You can even schedule those API tests to run on a cadence and catch cost regressions early. Want to follow along? Download Apidog and point a request at the Gemini endpoint before you wire it into production.
FAQ
How much does Gemini 3.6 Flash cost per 1M tokens? $1.50 for input and $7.50 for output. Context caching reads cost $0.15 per 1M, with a storage fee of $1.00 per 1M per hour.
Does the output price include thinking tokens? Yes. The reasoning the model does before it answers is billed at the $7.50 output rate. There’s no separate charge for thinking tokens, but they do count toward your output total, so more reasoning means a bigger bill.
Is there really a free tier? Yes, through Google AI Studio, and it’s rate-limited. It’s meant for prototyping and testing, not production. Google may use free-tier data to improve its products, so keep sensitive data on a paid key.
Is Gemini 3.6 Flash cheaper than 3.5 Flash? Yes. The output rate fell from $9.00 to $7.50, and 3.6 Flash uses about 17% fewer output tokens for the same task. Both effects stack, so the real-world output bill can drop around 31%.
When should I use Flash-Lite instead? When the task is simple and the volume is high. At $0.30 input and $2.50 output, Flash-Lite is roughly 70% cheaper for the same token counts, and it’s fast, but 3.6 Flash handles harder reasoning better.
Gemini 3.6 Flash is a rare pricing update where the number went down and the model got better at the same time. Budget it at $1.50 in and $7.50 out, remember that thinking tokens ride on the output rate, and lean on caching plus Flash-Lite routing to trim the rest. For historical context on how Gemini API costs have moved across releases, the Gemini 3.0 API cost breakdown is a useful anchor. Then measure your own calls in Apidog so the token counts you plan around are the ones you actually pay for.
