Google Releases Gemini 3.8 Flash: What the September 2026 Launch Means for Your Software

On September 2, 2026, Google released Gemini 3.8 Flash, the newest model in its faster, lower-cost Flash line. Google's Gemini API release notes list it as generally available under the model ID gemini-3.8-flash and describe it as the company's most intelligent Flash model, built for long-running software engineering tasks, autonomous agents and complex enterprise workflows.

For a business, the release matters less as a headline and more as a set of practical changes: a new default model for Google's agent tooling, introductory pricing with a fixed end date, and a handful of API settings that no longer work the way they did. If your custom software or web application calls the Gemini API, or you are deciding which model to build on, these are the details worth knowing.

This article separates what is available now from what is restricted or still in preview, and then sets out what to check in your own product.

What Google announced on September 2, 2026

Google's announcement on blog.google, published September 2, 2026, introduced two models. Gemini 3.8 Flash is the general-purpose model, focused on reasoning and coding. Gemini 3.8 Flash Cyber is a specialized variant for security work such as finding and patching software vulnerabilities.

The two are not equally available. According to the same post, Gemini 3.8 Flash is open to developers through Google AI Studio and Android Studio, to enterprises through Gemini Enterprise, and to consumers through the Gemini app, AI Mode in Google Search and Google Sheets. Gemini 3.8 Flash Cyber is limited to what Google calls trusted defenders, who apply through its new Fairwind Program. Unless your organization maintains critical infrastructure or widely used software, the Cyber variant is not something you can plan around today.

Google's post also reports benchmark results for both models. Those figures are the vendor's own, so treat them as a reason to run your own tests and not as a guarantee of how the model will perform on your data.

What the model can and cannot do

The Gemini 3.8 Flash model page lists an input limit of 1,048,576 tokens and an output limit of 65,536 tokens. It accepts text, images, video, audio and PDF files as input and produces text as output.

The same page lists the supported features: function calling, structured outputs, code execution, file search, grounding with Google Search and Google Maps, URL context, caching, the Batch API, and the Flex and Priority inference tiers. Computer use, where the model operates a browser or desktop interface, is listed as a preview feature and should not be treated as production-ready.

Three things are listed as unsupported: audio generation, image generation and the Live API. If your product needs spoken output or real-time voice conversations, Google directs you to separate models for those jobs. Gemini 3.8 Flash is a text-output model for reasoning, coding and tool use.

Pricing has an introductory period that ends on December 31, 2026

Google's announcement states that Gemini 3.8 Flash costs USD 0.75 per million input tokens and USD 3.75 per million output tokens at introductory rates through December 31, 2026. Google's latest model guide gives the standard rates that apply from January 1, 2027: USD 1.50 per million input tokens and USD 7.50 per million output tokens. The Gemini API pricing page shows the same two-stage pricing.

In plain terms, the published rate doubles at the start of 2027. Any cost estimate you build during the last quarter of 2026 should use the standard rate, not the introductory one, or your running costs will jump in January without any change in usage. Always confirm current figures on the pricing page before you commit to a budget, because rates and tiers can change.

API changes developers need to handle

Moving to Gemini 3.8 Flash is not always a one-line change. Google's latest model guide sets out a migration checklist, and several items affect existing code.

Sampling parameters are deprecated

The release notes entry for July 21, 2026 states that the sampling parameters temperature, top_p and top_k are now deprecated, and the migration guide tells developers to remove them from their request settings. Many applications set temperature to make output more or less predictable. If yours does, that behavior needs to be re-tested, and consistency should be enforced through clear instructions and structured outputs instead.

Thinking levels replace thinking budgets

The guide says the older thinking_budget setting is replaced by thinking_level, with three values: low, medium and high. Medium is the default. The guide also states that the minimal level is not supported on Gemini 3.8 Flash and returns an error, so code that requested minimal thinking on an earlier model will fail until it is updated.

Other items on the checklist

The guide notes that candidate_count is not supported on Gemini 3 models and recommends managing multi-turn conversations on the server side through the Interactions API. If you are coming from Gemini 2.5 or an early Gemini 3 model, it points to an earlier migration checklist that also needs to be worked through.

Gemini 2.5 access was restricted on September 18, 2026

A related change arrived two weeks later. The release notes entry for September 18, 2026 says that access to the Gemini 2.5 models is now limited to users who have actively used them in the past. Google states that these models are not deprecated and will continue to be served through the API, and that new projects should use Gemini 3.5 Flash-Lite or Gemini 3.8 Flash. The deprecations page shows no shutdown date for gemini-2.5-flash or gemini-2.5-pro.

So existing Gemini 2.5 integrations keep working for now, but the door is closed to new projects. That has a practical side effect: a new Google Cloud project created for staging, a new client or a new environment may not be able to call a 2.5 model even though your production project still can.

What this means for your business

The following is our reading of the announcements, offered as guidance and not as fact.

If you already run on an earlier Gemini model, nothing forces an immediate move. The pressure is gradual: the newest features and agent tools default to 3.8 Flash, and older models are being closed to new projects. Planning the upgrade on your own schedule is usually cheaper than doing it under a shutdown notice.

If you are starting a new AI feature, Gemini 3.8 Flash is the model Google is steering new work toward. That does not make it the right choice for every task. A simple classification or summarization job may run perfectly well on the cheaper Gemini 3.5 Flash-Lite, and the difference adds up at volume. Businesses comparing a simple assistant with a tool-using agent may find our guide to AI agents, chatbots and workflow automation useful before choosing a model at all.

Finally, a model that reasons for longer and calls more tools uses more tokens per task. A lower price per token does not automatically mean a lower bill.

What to do next

• Ask your developers which Gemini model IDs your software calls today, and in which Google Cloud projects.
• Search the codebase for temperature, top_p, top_k, thinking_budget and candidate_count, since these are the settings the migration guide flags.
• Run your own test set of real prompts against Gemini 3.8 Flash before switching, and compare quality, speed and token use with your current model.
• Re-forecast API costs using the standard rate that applies from January 1, 2027.
• Keep computer use and other preview features out of customer-facing workflows until Google marks them generally available.

Conclusion

Gemini 3.8 Flash has been generally available since September 2, 2026, with introductory pricing through December 31, 2026 and a short list of API changes that can break older integrations. Gemini 3.8 Flash Cyber is restricted to approved security teams, and Gemini 2.5 remains in service but closed to new users since September 18, 2026.

For most businesses the sensible response is a small, planned piece of work: an audit of current model usage, a test run and a revised cost estimate. Entrant Technologies builds websites, web applications, mobile apps and custom software, and if you would like help assessing a Gemini integration or migration, you can contact us to talk it through.