Gemini 3.8 Flash Guide: Pricing, Free Tier, Specs and Access [2026]

This page used to be our Gemini 3 Flash guide. That model is nine months and four Flash versions old now, and keeping a guide centred on it would have sent people to the wrong model, so we rewrote it around the one Google actually recommends today. If you landed here looking for Gemini 3 Flash, jump to what happened to it.

Gemini 3.8 Flash is the model Google wants most developers and paying Gemini users on right now. Google calls it its most intelligent workhorse model and claims gains over 3.7 Flash in software engineering, agentic tasks, and multi-step reasoning in specialized fields like finance and law. It launched alongside a security-focused sibling, Gemini 3.8 Flash Cyber, which is covered further down.

It outputs text. Gemini Omni Flash, Google's separate video model, shares the Flash name but does a different job, and we cover it in our Gemini Omni Flash guide. The naming is confusing, so check which one a tutorial means before you follow it.

The introductory rate matches what Google set for 3.7 Flash. The standard rate doubles both numbers, which means that whatever you spend in December, you should expect to spend twice as much in January for the same traffic. If you are building a cost model for next year, use $1.50 and $7.50, not the launch price. Google's docs also list Batch API and context caching support for Gemini 3 models. Those are the usual ways to bring a large bill down, and I would set them up before the rate doubles rather than after. For a side-by-side of what other providers charge, use our AI cost calculator or the task-by-task model guide.

In the Gemini app, 3.8 Flash is not free for now. Google's launch post lists Google AI Pro and Ultra subscribers as the people who get it in the Gemini app, AI Mode in Search, and Gemini in Google Sheets, and it does not say when free users will get it. App limits are compute-based. They depend on the model, the features you use, and how long the chat is, and they refresh every 5 hours up to a weekly cap. Google's help page says Pro gives 2x the usage of the free plan and Ultra gives 4x.

The API is a different story. New API accounts start on a free tier that covers certain models up to their free-tier rate limits. Google applies limits per project, not per API key, and daily request quotas reset at midnight Pacific time. What Google does not publish in its docs is a fixed free-tier number for 3.8 Flash. Your actual limits appear inside Google AI Studio and change with your usage tier, so look there before you plan a prototype around free requests.

That cutoff row is easy to skim past. Google says the model knows some areas up to March 2026 but others only up to January 2025, in line with the rest of the Gemini 3 family. If recent facts matter to your use case, turn on Search grounding rather than trusting the model's memory.

Google's launch post leans on three evaluations. On DeepSWE v1.1, a long-horizon software engineering benchmark, Google says 3.8 Flash beats most larger frontier models at a fraction of the cost. On Vals Finance Agent V2 and Harvey's Legal Agent Benchmark, it says 3.8 Flash beats 3.7 Flash and other frontier models. Google's evaluation methodology page adds that its DeepSWE scores for 3.8 Flash were self-computed with a mini-swe-agent harness at high thinking, and that its Terminal-Bench 2.1 scores for Gemini models are also self-computed.

You will notice there is no score table here. That is deliberate. We could not confirm the exact per-benchmark numbers from Google's pages at the time of this update, and we would rather show nothing than show a figure we cannot trace. The full table is on Google DeepMind's Gemini 3.8 Flash evaluation page. Keep in mind that these are Google's own runs, and Google notes its current results use updated evaluations that are not directly comparable with older Gemini model cards.

Gemini 3.8 Flash Cyber is a separate variant tuned for vulnerability detection and automated patching. Google gives trusted defenders, including government authorities and critical infrastructure operators, access through a new Fairwind Program. You cannot pick it as a general API option, and it has no public per-token pricing, so leave it out of any normal cost comparison. The earlier 3.5 Flash Cyber followed a similar limited-access pattern through a pilot program.

Read down that table and the shift is obvious. Google used to treat Flash as a cheap tier that shipped once a year. Now it updates it every few weeks, and it cut the launch price for 3.7 and 3.8. That is good for your bill, but it also means any model ID you hard-code will go stale quickly. Pin a version for stability, and put a reminder in your calendar to test the next one.

Gemini 3 Flash launched in December 2025 as a preview model with the ID gemini-3-flash-preview. It still has a model page in the Gemini API docs, and as of September 25, 2026 we could not find a shutdown date for it on Google's deprecations page. That does not make it safe to build on. Google says preview models can be deprecated with at least two weeks of notice, and it has already shut down other Gemini 3 previews, such as Gemini 3 Pro Preview in March 2026 and Gemini 3.1 Flash-Lite Preview in May 2026.

Google's models page now tells new projects to use Gemini 3.8 Flash or Gemini 3.5 Flash-Lite. If you still run gemini-3-flash-preview, move to gemini-3.8-flash. Check the price before you do. Even at the introductory rate, 3.8 Flash costs more per token than Gemini 3 Flash did at launch: $0.75 against $0.50 for input, and $3.75 against $3.00 for output. Re-run your cost numbers before you switch.

If you build agents or coding tools on Google Cloud, this is the model Google is tuning for that work, and it is GA on Agent Platform. Teams with long documents or large codebases get a 1M-token input limit, which handles a lot in one request. Google AI Pro and Ultra subscribers do not need to do anything, because it is already in the Gemini app, AI Mode, and Sheets.

The group I would worry about is anyone still on Gemini 3 Flash. Plan the migration now instead of waiting for a deprecation notice. With a preview model and a two-week minimum notice period, waiting is the riskier choice.

If you are choosing between Google and other providers for a specific task, that decision lives in our task-by-task model guide, which we keep updated as models change. For agentic coding tools more broadly, see our Claude Code guide.

• Google: Introducing Gemini 3.8 Flash and 3.8 Flash Cyber
• Google Gemini API: Gemini 3.8 Flash model page
• Google Gemini API: What's new in Gemini 3.8 Flash
• Google Gemini API: Pricing
• Google Gemini API: Rate limits
• Google Gemini API: Models
• Google Gemini API: Deprecations
• Google Gemini API: Gemini 3 developer guide
• Google DeepMind: Gemini 3.8 Flash model card
• Google DeepMind: Gemini 3.8 Flash evaluation methodology and results
• Google Cloud: Gemini 3.8 Flash on Gemini Enterprise Agent Platform
• Google: Gemini 3.7 Flash announcement
• Google: Introducing Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber
• Google: Gemini 3.5 announcement
• Google Gemini Apps Help: Limits and upgrades for Google AI subscribers

How much does Gemini 3.8 Flash cost?

In the Gemini API, Gemini 3.8 Flash costs $0.75 per million input tokens and $3.75 per million output tokens at its introductory rate, which runs through December 31, 2026. From January 1, 2027, Google lists a standard rate of $1.50 per million input tokens and $7.50 per million output tokens.

Is Gemini 3.8 Flash free?

Not in the Gemini app for now. Google's launch post makes 3.8 Flash available to Google AI Pro and Ultra subscribers in the Gemini app, AI Mode in Search, and Gemini in Google Sheets. For developers, new Gemini API accounts start on a free tier, but Google shows the exact per-model limits in Google AI Studio rather than in its public docs, so check your project's limits there before you plan around free usage.

What is the context window of Gemini 3.8 Flash?

Gemini 3.8 Flash accepts up to 1 million input tokens and can return up to 64,000 output tokens. Google lists its knowledge cutoff as March 2026, with a caveat that in some domains its knowledge stops at January 2025.

Is Gemini 3 Flash shut down?

We could not find a shutdown date for gemini-3-flash-preview on Google's deprecations page as of September 25, 2026. It is a preview model, and Google says preview models can be deprecated with at least two weeks of notice. Google's models page now points new projects to Gemini 3.8 Flash or Gemini 3.5 Flash-Lite, so treat Gemini 3 Flash as legacy and plan a migration.

What is the difference between Gemini 3.8 Flash and Gemini 3.8 Flash Cyber?

Gemini 3.8 Flash is the general model anyone can use through the API or a paid Google AI plan. Gemini 3.8 Flash Cyber is a separate variant tuned for finding and patching software vulnerabilities, and Google only offers it to trusted defenders, such as government authorities and critical infrastructure operators, through its Fairwind Program.

Written by

Paras Tiwari

Founder, Spectrum AI Labs

Founder of Spectrum AI Labs — testing AI tools and models, and writing up what actually ships.

More about Paras →