Google Gemini 3.8 Flash is a production-ready AI model built for long-horizon software engineering, autonomous agents and complex workflows, with a 1M-token context window and a broad set of built-in tools.

The Short Version

Google has launched Gemini 3.8 Flash, a new production-ready model built for long-horizon software engineering, autonomous agents and complex multi-step workflows. Google describes it as its most intelligent Flash model yet. It combines a 1 million-token context window with built-in tools for code execution, file search, function calling, Search grounding, Google Maps grounding, structured outputs and computer use in preview.

Flash Is Becoming an Agentic Workhorse

Gemini 3.8 Flash pushes Google’s Flash line deeper into agentic work. The model is designed for tasks that continue across many steps rather than ending after one prompt and one answer. That makes it a natural fit for coding agents, research workflows, application assistants and systems that repeatedly plan, call tools, inspect results and continue working.

Built for Long-Horizon Software Engineering

Google positions long-horizon software engineering as one of the core strengths of Gemini 3.8 Flash. The model is designed to stay engaged across larger coding jobs that involve several files, multiple tool calls and a sequence of related decisions. For developers, this points toward AI that can participate in a broader slice of the software workflow instead of focusing only on isolated code snippets.

Autonomous Agents Are a First-Class Use Case

Gemini 3.8 Flash is also engineered for autonomous agents. These systems can break a goal into steps, select tools, inspect their own progress and continue until a larger task is complete. Google is clearly treating agent orchestration as a central part of the model’s identity, which makes 3.8 Flash especially relevant for the next generation of AI-powered applications.

A 1 Million-Token Context Window

The model supports an input context window of 1,048,576 tokens. That gives developers a large working space for source code, documentation, long conversations, reports, project files and other context-heavy material. A large context window becomes particularly useful when an agent needs to keep many connected pieces of information available throughout a multi-step workflow.

Up to 65,536 Output Tokens

Gemini 3.8 Flash supports up to 65,536 output tokens. That creates room for substantial generated work such as detailed technical reports, large code changes, structured plans and long-form analysis. Combined with the 1 million-token input window, the model is set up for workflows where both the source material and the resulting output can be large.

Multimodal Inputs Are Built In

Google lists text, images, video, audio and PDFs as supported input types for Gemini 3.8 Flash. That means one agent can reason across several kinds of information without every workflow being reduced to plain text first. For applications that combine documents, screenshots, media and written instructions, this gives the model a much richer view of the task.

Code Execution Can Be Part of the Workflow

Gemini 3.8 Flash supports code execution as a built-in capability. This gives an application a way to move from reasoning about a technical task to running computations or validating generated logic inside the workflow. In agentic systems, that connection between thinking and execution can make the model more useful for data tasks, technical analysis and software development.

Computer Use Is Available in Preview

Google also lists computer use as a preview capability for Gemini 3.8 Flash. Computer use allows an AI system to interact with graphical software and interfaces as part of a larger task. Combined with planning and tool use, it points toward applications that can move through real digital workflows rather than operating only inside a chat box.

Function Calling Supports Connected Applications

Function calling is supported, giving developers a direct way to connect Gemini 3.8 Flash with application logic and external tools. An agent can decide when a function is needed, request the appropriate action and use the returned information as part of the next step. This is one of the core building blocks behind practical AI assistants that can work with live systems.

File Search Helps Agents Work Across Project Material

File search is another built-in capability. For agents working with project documents, technical references or large collections of material, file search can provide a direct retrieval layer inside the model workflow. That helps turn a large document set into active working context instead of something a user has to manually copy into prompts.

Search Grounding Connects the Model to Fresh Information

Gemini 3.8 Flash supports grounding with Google Search. For applications that need current public information, this gives the model a path to bring fresh web context into its reasoning process. Grounded search is especially useful for research agents, information assistants and applications that need to combine model reasoning with up-to-date external material.

Google Maps Grounding Opens Another Data Layer

The model also supports grounding with Google Maps. That creates interesting opportunities for applications involving places, travel, business discovery and geographic context. When combined with agentic planning, Maps grounding can become one more specialized information source that an AI system can call as it works through a real-world task.

Structured Outputs Help Applications Stay Organized

Structured outputs are supported in Gemini 3.8 Flash. This allows developers to request predictable machine-readable response formats for application workflows. Agents often need to pass results from one step to another, so structured outputs can make it easier to turn model responses into data that software can reliably process.

Thinking Levels Can Be Tuned

Google provides low, medium and high thinking levels for Gemini 3.8 Flash. This gives developers a way to shape how much reasoning effort the model applies to different classes of tasks. A lightweight interaction and a complex multi-step engineering job can therefore use different reasoning settings while staying on the same model.

URL Context Expands What an Agent Can Work With

URL context is supported as well. This gives applications another convenient way to bring external material into a task without manually reformatting everything into a prompt. For research, documentation and web-based workflows, direct URL context can help an agent move more smoothly between source material and generated work.

Caching Supports Repeated Workflows

Gemini 3.8 Flash supports caching, which is useful when an application repeatedly works with the same large body of context. Long-running agents often revisit project instructions, codebases or reference material across several steps. Caching gives developers another tool for building persistent workflows around substantial shared context.

Gemini 3.8 Flash Is Generally Available

Google lists Gemini 3.8 Flash as generally available and ready for production use. The stable model ID is gemini-3.8-flash. That production status matters for developers because Google is positioning the model for real applications, agent systems and enterprise workflows now.

Google AI Studio Is One Starting Point

Developers can start working with Gemini 3.8 Flash through Google AI Studio and the Gemini API. AI Studio gives builders a direct environment for testing prompts and model behavior, while the API provides the path into custom applications. This makes the new Flash model immediately accessible to both rapid experimentation and production development.

Gemini Enterprise Adds a Business Deployment Path

Google is also making Gemini 3.8 Flash available through Gemini Enterprise. That gives organizations a deployment path for agentic and multi-step workflows using Google’s enterprise AI platform. The same model can therefore sit behind individual developer experiments and larger managed applications.

Antigravity Uses Gemini 3.8 Flash by Default

Google’s documentation says the Antigravity agent and Antigravity SDK use Gemini 3.8 Flash by default. That is a strong signal about where Google sees the model fitting: at the center of agent workflows that need planning, tool orchestration and sustained work across multiple steps.

The Gemini App Is Part of the Rollout

For consumers, Google says Gemini 3.8 Flash is available to Google AI Pro and Ultra subscribers in the Gemini app. This gives the model a consumer-facing route alongside the developer and enterprise paths, bringing its new reasoning and agentic capabilities into Google’s broader AI ecosystem.

AI Mode in Google Search Also Gets 3.8 Flash

Google is also bringing Gemini 3.8 Flash to AI Mode in Google Search for Google AI Pro and Ultra subscribers. That puts the model’s reasoning capabilities directly inside one of Google’s most widely used information products and shows how the Flash line is becoming part of more than standalone AI tools.

Gemini in Sheets Is Included

Gemini 3.8 Flash is also available in Gemini in Google Sheets for eligible subscribers. Spreadsheet work is a natural place for reasoning, structured data and multi-step assistance to come together. The inclusion of Sheets shows how Google is extending the model into everyday productivity workflows as well.

Google Demonstrated an Entire 3D Game From One Prompt

One of Google’s launch demonstrations shows Gemini 3.8 Flash building a functional 3D game from a simple prompt using a looping instruction inside Google Antigravity. The demo combines puzzles, environmental storytelling and generated textures, providing a vivid example of the model working through a creative software task rather than producing only a short piece of code.

A DOS-Style Google Maps Demo Shows Its Range

Google also demonstrated Gemini 3.8 Flash creating a playable DOS-style version of Google Maps from a single prompt in Antigravity. The result includes locations, directions and Street View. It is a playful showcase, but it also demonstrates the model’s ability to combine interface work, application logic and connected information inside one generated experience.

Hardware Anatomy Turns Device Teardowns Into Interactive 3D

Another Google demo, Hardware Anatomy, uses Gemini 3.8 Flash in Google AI Studio to generate interactive Three.js visualizations of hardware devices. The experience breaks a device into layers that can be explored with a deconstruction slider. It is a strong example of the model combining technical understanding with visual software generation.

Introductory API Pricing Starts at $0.75 Per Million Input Tokens

Google lists introductory Gemini 3.8 Flash pricing at $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026. That pricing gives developers a clear entry point for experimenting with long-context, tool-using and agentic applications on the new model.

Why This Launch Matters

Gemini 3.8 Flash brings several trends together in one model: large context, multimodal input, built-in tools, configurable reasoning, computer use and a strong focus on autonomous agents. The important shift is that Flash is becoming a workhorse for applications that can keep going, use tools and complete larger digital tasks.

The Upgrade Feeling

The most interesting thing about Gemini 3.8 Flash is how much of the modern AI workflow now fits inside one fast model tier. A developer can give it a large project context, connect functions and search, let it execute code, bring in files, ground its work with live information and build an agent around the same model. That feels less like a simple model update and more like a new foundation for software that can actually work through a task.

Generative Video Is Starting to Behave More Like an Editing Workflow

Generative video began with a simple interaction.

Write a prompt.

Generate a clip.

Choose whether to keep it.

That model is still useful, but it is no longer the whole workflow.

Gemini Omni 1.1 Flash adds controls that move generation closer to the structure of video editing.

A creator can define the first frame.

Define the last frame.

Generate the movement between them.

Extend an existing scene.

Reuse part of an earlier video as reference context.

Create a low-resolution draft.

Then render or upscale the chosen result at a higher resolution.

Those steps look less like one isolated generation and more like a sequence of editorial decisions.

Google released Omni 1.1 Flash on August 27, 2026 and made the new capabilities available through the Gemini API, Google AI Studio, Gemini Enterprise Agent Platform and Google Flow.

The model still generates video.

The difference is how much of the generation process can now be shaped around a planned sequence.

The creator is beginning to specify not only what should appear.

They can also specify where the shot begins, where it ends and how the sequence continues.

First and Last Frames Create a Keyframe-Like Control Point

One of the clearest additions is first-and-last-frame interpolation.

Gemini Omni 1.1 Flash can take one image as the starting frame and another image as the final frame.

The model then generates continuous video between them.

That changes the creative problem.

Instead of asking the model to invent both the start and the end of a shot, the creator can define those endpoints directly.

The generated part becomes the transition.

Google gives examples including camera orbits, zoom transitions and seamless loops.

The API documentation exposes the same structure with media-role tags such as FIRST_FRAME and LAST_FRAME.

That makes the control explicit.

The first image is not only a reference.

It is the opening state of the shot.

The last image is not only style guidance.

It is the target state the generated sequence should reach.

This resembles a basic editing or animation idea: establish key points in time, then create the movement between them.

The model is still doing the generation.

The creator has more control over the temporal boundaries of the shot.

A Camera Move Can Now Be Defined by Its Destination

Camera direction becomes more specific when the destination is known.

A text prompt such as “zoom in” describes motion.

A first-and-last-frame pair can describe the visual result that motion should connect.

That gives the model more information about the intended path.

Google demonstrates this with continuous camera moves such as whip pans, orbits and zooms.

The creator can choose the opening composition.

Choose the final composition.

Then describe how the camera should travel between them.

This is different from asking for a complete shot from text alone.

The final composition is already fixed by the input.

The generation problem is narrowed to continuity, motion and transition.

For production workflows, that can be useful when the next shot needs to land on a particular framing.

A video can begin on a wide composition and end on a close-up.

It can start outside a room and end inside.

It can begin with one subject position and end with another.

The model is not only deciding what a camera move might look like.

It is generating toward a defined visual destination.

The Same Frame Can Define Both Ends of a Seamless Loop

The first-and-last-frame system also creates a direct looping workflow.

Google’s developer documentation shows that the same image can be used as both the first frame and the last frame.

The generated motion then begins from that image and returns to it.

That gives creators a structured way to build looping clips.

A product can rotate and return to its initial orientation.

A scene can move through an environmental animation and come back to the same composition.

A background can cycle while preserving a continuous boundary between the end and the beginning.

Looping video has traditionally required careful alignment between the last frame and the first frame.

Here, the endpoint is defined before generation.

The model receives the loop condition as part of the input structure.

That is another example of generative video becoming more timeline-aware.

The creator is not asking only for content.

They are defining a temporal relationship between the beginning and the end of the clip.

Scene Extension Turns One Clip Into the Beginning of a Longer Sequence

Omni 1.1 Flash also adds scene extension.

The model can take an existing video and continue generating from where it ends.

Google’s current documentation says extensions are generated in 10-second increments and can continue to a cumulative length of up to 40 seconds.

The model uses the preceding video as context for the continuation.

That changes how longer sequences can be built.

A creator can generate an initial scene.

Review it.

Then decide what should happen next.

The next prompt can continue the same location, introduce another action, change the camera movement or move the story into another connected scene.

The previous clip becomes input to the next generation step.

This creates a sequential workflow.

Generate.

Inspect.

Extend.

Inspect again.

Continue.

The model is no longer limited to producing one independent short clip at a time.

The previous shot can become the context for the next part of the sequence.

Ten Seconds of Prior Context Helps the Extension See More of the Scene

The amount of prior video available to the model matters when a scene continues.

Google says Omni 1.1 Flash can analyze up to 10 seconds of previous video context when generating an extension.

The company contrasts that with earlier models that referenced only the final second.

A larger context window gives the extension access to more of what happened before the cut point.

That can include movement direction.

Character position.

Camera motion.

The visual layout of the environment.

Objects introduced earlier in the shot.

Audio or dialogue already present in the sequence.

The extension can therefore be conditioned on a longer slice of the previous scene.

This is a timeline concept again.

The next generated segment is not built from one frozen endpoint alone.

It can look back across several seconds of preceding motion before deciding how the continuation should unfold.

For longer-form generation, that creates a stronger link between one segment and the next.

The clip becomes history for the following generation step.

Extension Prompts Can Direct Both Continuity and Change

Scene extension is not limited to “continue exactly as before.”

The prompt can describe what should remain and what should change.

Google’s documentation gives examples of continuing the same characters, changing the music, introducing a scene cut or directing a new camera move.

That gives the creator a way to edit the future of the clip.

The existing video defines the past.

The prompt defines the next event.

The model generates the transition between those two states.

This structure fits narrative work especially well.

A conversation can continue.

A camera can pull back to reveal more of the location.

A character can move into a new part of the scene.

The audio can continue or change.

The next extension can then build on that result.

The process becomes iterative rather than one-shot.

The creator can make decisions at each stage of the sequence instead of having to specify the entire finished video in one prompt.

360p Drafts Add a Preview Stage Before the Final Render

Another production-oriented change is the 360p draft mode.

Google says Omni 1.1 Flash can generate 360p previews up to 60 percent faster than its standard 720p output based on system throughput.

Google also says the 360p option costs one third of the standard 720p generation price in the current API pricing structure.

The purpose is iteration.

A creator may need several versions before choosing a shot.

One version changes the camera path.

Another changes timing.

Another changes the subject position.

Another changes the ending frame.

Generating every experiment at final resolution uses more time and compute than the creative decision requires.

A draft stage separates composition from finishing.

Generate a lightweight preview.

Compare alternatives.

Select the version that works.

Then move the selected shot into the higher-resolution stage.

That is a familiar pattern in editing, visual effects and 3D production.

The early decision does not require the final render.

Omni 1.1 Flash now gives generative video the same kind of two-stage workflow.

Drafting Makes Variation Testing More Structured

Low-resolution previews become more useful when they are treated as controlled experiments.

Google describes a “Draft Room” concept where creators can generate several 360p versions and change one variable at a time.

That might be camera direction.

Timing.

Lighting.

Character position.

The final frame.

The extension prompt.

The creator can then compare the variations side by side.

This changes prompting from repeated guessing into a more structured process.

Keep most of the shot fixed.

Change one parameter.

Observe the result.

Then decide which direction to continue.

That resembles ordinary creative iteration.

An editor tests several cuts.

A designer compares versions.

A photographer changes one exposure variable.

A 3D artist renders a low-quality preview before committing to final settings.

Generative video can now support a similar loop.

The model remains probabilistic, but the workflow around it can become more systematic.

The draft is not the final asset.

It is evidence for the next creative decision.

1080p and 4K Create a Separate Finishing Stage

Once the shot has been selected, Omni 1.1 Flash supports higher-resolution output.

Google lists 1080p and 4K as available output resolutions for the current model.

That creates a clear separation between preview and final delivery.

The 360p generation can be used for fast iteration.

The selected result can then move to a higher-resolution output for production use.

This matters because resolution becomes a workflow variable rather than something fixed at the beginning.

The creator can spend compute where it matters.

Early experimentation can stay lightweight.

The chosen clip can receive the higher-resolution treatment.

Google Flow uses the same general idea in its creative interface, allowing users to draft at lower resolution and move selected work into higher-quality output.

The result is a more familiar production sequence.

Plan.

Preview.

Choose.

Finish.

The generation model is becoming one stage inside that sequence rather than the entire sequence by itself.

Video References Add Motion and Character Context to the Prompt

Omni 1.1 Flash can also accept short video references.

Google says the model can use up to three seconds of reference video when building a new scene.

That gives the prompt another type of context.

An image reference can show how a subject looks.

A video reference can also show movement.

A dance.

A gesture.

A camera behavior.

A character performance.

The model can then use that reference while generating a different scene.

Google demonstrates a workflow where several characters are assigned movements derived from reference videos.

The reference is not necessarily the video being edited.

It can be guidance.

That expands the role of input media.

Text describes intention.

Images can define appearance or keyframes.

Video can provide motion reference.

An existing generated clip can provide extension context.

The creator can combine those media types to define different parts of the shot.

The prompt becomes a composition of roles rather than one block of text.

Media Roles Make Multimodal Prompting More Explicit

Google’s API documentation now gives uploaded media explicit roles.

FIRST_FRAME.

LAST_FRAME.

IMAGE_REF.

VIDEO_REF.

PREVIOUS_VIDEO.

Those labels matter because the same media file can be used in different ways.

An image used as FIRST_FRAME is the literal opening frame.

The same image used as IMAGE_REF provides guidance without becoming the opening frame.

A video used as PREVIOUS_VIDEO becomes the sequence being extended.

A video used as VIDEO_REF becomes reference context.

That distinction makes multimodal prompting more like a production specification.

The creator does not only upload assets.

They define what job each asset performs.

This is similar to the way an editing project separates source media, timeline clips, reference footage and output targets.

The model receives both the media and the role.

That reduces ambiguity at the interface level.

The creative system knows which asset is supposed to be a boundary, which is supposed to be reference material and which is supposed to be the existing sequence being continued.

Conversational Editing Keeps the Same Video Inside an Ongoing Interaction

Gemini Omni was designed around conversational video generation and editing.

The Interactions API allows a developer to keep working with the same creative sequence across turns.

A user can generate a clip.

Then ask for an edit.

Then extend it.

Then change another part of the scene.

The interaction history becomes part of the workflow.

This is different from exporting a clip and starting a completely new generation every time.

The model can remain inside an ongoing creative session.

Google introduced Omni around the idea of combining text, images, audio and video as input and refining video through conversation.

Omni 1.1 Flash adds more structured controls to that conversation.

The prompt still matters.

Now the creator can combine the prompt with keyframes, reference media, previous video and resolution choices.

Conversation becomes the control layer around those assets.

The editing session is therefore both natural-language driven and media-structured.

Google Flow Turns the New Controls Into a Creator Interface

The same features are appearing in Google Flow.

Google announced the Omni 1.1 Flash controls in Flow on August 27, 2026.

Creators can define start and end frames.

Generate 360p drafts.

Move selected work into higher-resolution output.

Continue shaping video inside the Flow environment.

This matters because the capabilities are not limited to API developers.

The same model concepts are being translated into a visual creative tool.

A creator can think in terms of shots and transitions instead of request objects and API fields.

The developer sees FIRST_FRAME and LAST_FRAME.

The creator sees start frame and end frame.

The underlying idea is the same.

The model is receiving timeline boundaries.

This is how a generative capability becomes part of an editing workflow.

The technical control appears first as model functionality.

The creative application turns that control into an interface a person can use repeatedly.

The Model Can Sit Inside Existing Creative Tools

Google is also positioning Omni Flash as a model that can be integrated into other creative software.

Google’s August announcement cites Adobe Firefly, Figma Weave and Runway among tools using or integrating Gemini Omni Flash.

That matters for workflow design because the generative model does not need to become the entire editing application.

It can become one capability inside a larger creative environment.

The host application can manage projects.

Assets.

Versions.

Canvas organization.

Collaboration.

Export.

Omni can provide generation and editing operations inside that system.

This follows the same architectural pattern seen across AI software.

The model supplies a capability.

The application supplies the workflow around it.

For video production, that means generative controls can live beside conventional editing, design and asset-management tools rather than requiring a separate creative process.

Generative Video Is Becoming a Sequence of Controlled Decisions

The important change in Omni 1.1 Flash is not one isolated feature.

It is the combination.

First frame.

Last frame.

Scene extension.

Ten seconds of prior video context.

Reference video.

Media roles.

360p previews.

1080p and 4K output.

Conversational editing.

Each feature controls a different part of the process.

Where the shot starts.

Where it ends.

What came before it.

What movement or character should be referenced.

How quickly a draft should be generated.

When the output should move to finishing resolution.

Together, those controls create a sequence of decisions around the generated video.

That is what makes the workflow feel closer to editing.

The creator does not surrender the entire timeline to one prompt.

The model generates inside boundaries the creator can define at several stages.

Generative video remains generative.

The workflow around it is becoming more deliberate.

That is the upgrade.