Gemini 3.7 Flash: New Features and Key Upgrades

Google released Gemini 3.7 Flash on August 13, 2026, only three weeks after Gemini 3.6 Flash. Rather than introducing a completely different model family, the new release builds on 3.6 with algorithmic improvements aimed at making Flash more capable in real-world workflows.

Google is positioning Gemini 3.7 Flash particularly around coding and agents, but the update reaches beyond software development. Improvements also show up in web development, complex document understanding, computer use, long-context tasks, and enterprise automation. At the same time, the model keeps the large context window and multimodal input support already associated with the Gemini 3 family.

So what is Gemini 3.7 Flash, what actually changed from 3.6, and where might the new model be useful?

What Is Gemini 3.7 Flash?

Gemini 3.7 Flash is the latest iteration of Google’s Gemini 3 series of natively multimodal reasoning models. Google describes it as its most capable Flash workhorse yet for coding and agents, while the DeepMind model card says it introduces algorithmic improvements to the reasoning foundation inherited from Gemini 3.6 Flash.

The model supports:

Text, images, video, audio, and PDFs as input
Text as output
Up to 1,048,576 input tokens
Up to 65,536 output tokens
Low, medium, and high thinking levels
Function calling
Code execution
File search
Search grounding
Structured outputs
URL context
Caching
Computer use in preview

Its stable API model ID is gemini-3.7-flash.

This combination makes Gemini 3.7 Flash relevant to much more than ordinary chat. It can work with long documents, visual references, video, large code contexts, and tool-enabled workflows while retaining the speed-and-efficiency focus of the Flash family.

Gemini 3.7 Flash vs Gemini 3.6 Flash: What Actually Changed?

Gemini 3.7 Flash is based on Gemini 3.6 Flash, so this release is better understood as a refinement of the existing Flash foundation rather than a ground-up redesign.

The biggest change is not a new input format or a dramatically larger context window. Instead, Google has focused on how reliably the model handles work that extends beyond a single response.

According to Google, 3.7 Flash puts more effort into multi-step planning and tool calls, follows instructions more closely, and adapts better when it encounters roadblocks. These changes are intended to reduce retries and the amount of manual oversight required in engineering workflows.

That makes the 3.6-to-3.7 update less about adding one headline feature and more about making the existing capabilities work better together. The benchmark results show where those improvements are most visible.

Gemini 3.7 Flash Benchmarks at a Glance

Google DeepMind published comparisons between Gemini 3.7 Flash and Gemini 3.6 Flash across coding, agents, knowledge work, multimodal understanding, and long-context evaluation.

Benchmark	Gemini 3.7 Flash	Gemini 3.6 Flash
Artificial Analysis Intelligence Index	56	52
FrontierCode 1.1 Main	43.6%	34.4%
DeepSWE v1.1	65.3%	48.6%
Code Arena — Web Development	1588 Elo	1538
Terminal-bench 2.1	85.8%	78.0%
AutomationBench	30.4%	17.0%
GDP.pdf	34.0%	22.0%
OSWorld-2.0	47.9%	33.8%
LVBench	85.4%	84.2%
128K Long Context	97.0%	91.8%

The largest jumps are concentrated in tasks that require sustained execution. DeepSWE, for example, evaluates long-horizon software engineering, while AutomationBench focuses on enterprise workflow automation and OSWorld-2.0 evaluates agentic computer use.

The results are not uniformly higher across every test. On CharXiv, which evaluates information synthesis from complex charts, Gemini 3.7 Flash scored slightly below 3.6 Flash both without tools and with tools.

That makes it more accurate to describe 3.7 as a targeted improvement rather than a universal upgrade across every benchmark.

Gemini 3.7 Flash Pricing

Gemini 3.7 Flash launched with introductory API pricing of:

$0.75 per 1 million input tokens
$3.75 per 1 million output tokens

The introductory rate runs through December 31, 2026. Starting January 1, 2027, the pricing is scheduled to become:

$1.50 per 1 million input tokens
$7.50 per 1 million output tokens

Google describes the launch rate as half the original Gemini 3.6 Flash cost. However, the same promotional rate has also been extended to Gemini 3.6 Flash.

As a result, 3.6 and 3.7 currently share the same introductory token pricing.

That distinction is important when comparing the two models today. Gemini 3.7 Flash is not currently half the price of 3.6 Flash; the 50% comparison refers to the earlier 3.6 pricing.

How to Access Gemini 3.7 Flash

Developers can use Gemini 3.7 Flash through Google AI Studio and the Gemini API. It is also available through Google Antigravity, while enterprise users can access it through Gemini Enterprise products. Gemini Spark also uses the model for eligible Google AI Pro and Ultra subscribers.

For Gemini API integrations, the stable model code is:

gemini-3.7-flash

Google AI Studio provides a straightforward way to test prompts and explore the model before integrating it into an application.

What Can You Use Gemini 3.7 Flash For?

The technical improvements are easier to understand when translated into actual workflows rather than benchmark categories.

1. Maintain and Debug Existing Codebases

Instead of using the model only to generate isolated code snippets, developers can give Gemini 3.7 Flash a broader development task.

For example, it could inspect several related files, trace a bug through an application, propose changes, run code through supported tools, and continue working after an initial attempt fails.

That makes it more relevant to maintenance and debugging workflows where understanding how multiple parts of a project fit together matters as much as writing new code.

2. Turn Visual References Into Working Interfaces

Because Gemini 3.7 Flash accepts images, developers can combine a screenshot, mockup, or other visual reference with instructions for a web interface.

A workflow might involve giving the model a reference landing page and asking it to reproduce the structure as a responsive frontend, then iterating on specific components.

Google says the model can generate more functional layouts and feature-complete applications in fewer prompts, while showing stronger adherence to screenshots, images, and design systems.

3. Power Tool-Using AI Agents

Gemini 3.7 Flash can serve as the reasoning layer behind agents that need to interact with outside systems.

A business agent, for example, might search for information, process several files, call an internal tool, organize the results, and update a document. A research agent could gather information from multiple sources and structure the findings into a report.

The important part is that the task consists of a sequence of actions rather than one prompt followed by one answer.

4. Work With Large Reports and PDFs

The combination of PDF input and a 1M-token context window opens up workflows involving documents that would be inconvenient to break into dozens of separate prompts.

Possible examples include:

Annual reports
Research papers
Technical manuals
Legal documents
Financial reports
Internal company documentation

A user could ask questions across different sections of the same report, compare information found on separate pages, extract important figures, or turn a dense document into a more structured summary.

Google has demonstrated a similar workflow by turning a complex annual report into an interactive data experience containing charts and aggregated insights.

5. Analyze Images, Audio, and Video

Gemini 3.7 Flash can accept multimedia content directly, making it useful when the information a user needs is not purely textual.

For example, a video could be analyzed for:

The sequence of events
Actions performed by people or objects
Scene changes
Spoken information
Visual details
Continuity issues

Images can similarly be used for design interpretation, document understanding, visual analysis, or as references within coding tasks.

This is particularly useful when a workflow needs to understand existing media rather than generate new media.

Can Gemini 3.7 Flash Generate Images or Videos?

No. Gemini 3.7 Flash is multimodal, but its current output modality is text.

Google’s Gemini API documentation lists image generation and audio generation as unsupported for this model. Video generation is likewise not one of its direct output capabilities.

So Gemini 3.7 Flash can handle workflows such as:

Image → analysis

Video → analysis

Audio → analysis

PDF → analysis

But it is not the model used for:

Prompt → generated image

Prompt → generated video

Google has separate generative media models, including Nano Banana models for images and Veo models for video.

Gemini 3.7 Flash can still participate in a larger creative workflow—for example, by analyzing a reference, preparing a generation prompt, or coordinating another model—but it is not itself an image or video generator.

Should You Upgrade From Gemini 3.6 Flash?

For a new project, Gemini 3.7 Flash is a sensible version to test first. It is newer, currently carries the same promotional token price as 3.6 Flash, and shows stronger results across many of Google’s published evaluations.

For an existing Gemini 3.6 Flash application, the answer is less automatic.

A production workflow may already have prompts, tools, retry logic, latency targets, and output handling tuned around 3.6. Switching models simply because a benchmark number is higher does not guarantee that the complete application will perform better.

A more useful comparison is to run the same real workload through both models and measure factors such as:

Successful task completion
Number of retries
Tool-call failures
Latency
Output token use
Prompt compatibility
Total cost per completed task

If 3.7 completes the same workflow more reliably without creating new problems elsewhere, the upgrade becomes easier to justify.

Gemini 3.7 Flash FAQ

When was Gemini 3.7 Flash released?

Google introduced Gemini 3.7 Flash on August 13, 2026, three weeks after Gemini 3.6 Flash.

What is the Gemini 3.7 Flash API model ID?

The stable model ID is gemini-3.7-flash.

What is the Gemini 3.7 Flash context window?

It supports up to 1,048,576 input tokens and 65,536 output tokens.

What input types does Gemini 3.7 Flash support?

Gemini 3.7 Flash accepts text, images, video, audio, and PDFs. Its output is text.

Can Gemini 3.7 Flash generate images?

No. Google currently lists image generation as unsupported for Gemini 3.7 Flash.

Can Gemini 3.7 Flash analyze videos?

Yes. Video is one of its supported input types, so the model can reason about information contained in uploaded video.

How much does Gemini 3.7 Flash cost?

Through December 31, 2026, introductory pricing is $0.75 per million input tokens and $3.75 per million output tokens. Pricing is scheduled to become $1.50 and $7.50 respectively on January 1, 2027.

What is Gemini 3.7 Flash’s knowledge cutoff?

Google DeepMind lists March 2026 as the knowledge cutoff, while noting that knowledge in some domains may effectively be limited to January 2025.

Final Thoughts

Gemini 3.7 Flash is a relatively small version jump, but it arrives at an interesting moment for Google’s Gemini lineup.

Instead of introducing a completely different kind of model, Google is iterating quickly on Flash and pushing it further into tasks that were once more closely associated with larger, heavier models. The release came only three weeks after Gemini 3.6 Flash, reinforcing how quickly Google is refining this part of the Gemini family.

For users, the practical value of Gemini 3.7 Flash will become clearer as it is tested across real applications. For now, the release shows that Google is treating Flash not simply as a lightweight alternative, but as an increasingly important part of its broader Gemini strategy.

Gemini 3.7 Flash Is Here: Google’s New Flash Model Explained