Google Gemini combines a billion-user AI assistant with the new Gemini 3.7 Flash model

Google Gemini has grown far beyond a chatbot. The Gemini name now covers Google’s consumer AI assistant, a family of multimodal AI models, developer APIs and AI features integrated across Google products.

The distinction matters because “Gemini” does not refer to one fixed model. Someone using the Gemini app, a developer calling the Gemini API and a business deploying Gemini Enterprise may all be using different Gemini models and features.

Google said on 11 August 2026 that the Gemini app had passed one billion monthly active users. Just two days later, Google made Gemini 3.7 Flash generally available, giving the platform a newer model focused on reasoning, coding, multimodal work and agentic tasks.

TechnologyBlog.co.za has not independently benchmarked Gemini 3.7 Flash or tested every Gemini feature. This article therefore focuses on Google’s current specifications, documented capabilities and important limitations.

What is Google Gemini?

Gemini operates at several different levels.

For ordinary users, the Gemini app provides an AI assistant through the web, Android, iOS and other supported Google platforms. Users can ask questions, work with documents, generate content, conduct research and interact using text, images, audio and other media.

For developers, Google provides Gemini models through the Gemini API and Google AI Studio. Businesses can also access Gemini through Google’s enterprise products and agent platforms.

Google DeepMind develops the underlying Gemini model family, while Google integrates those models into consumer and business services.

This means the model running a particular Gemini feature can change as Google updates the service.

Gemini 3.7 Flash is the newest major Flash model

Google released Gemini 3.7 Flash on 13 August 2026 as a generally available model.

It builds on Gemini 3.6 Flash and introduces algorithmic improvements to the model’s reasoning foundation. Google positions it as a workhorse model for coding, agents, knowledge work and complex multimodal tasks.

Gemini 3.7 Flash accepts:

• Text
• Images
• Video
• Audio
• PDF documents

Its standard API output is text.

The model supports a 1,048,576-token input limit and a maximum output of 65,536 tokens. Developers can also choose low, medium or high thinking levels to balance reasoning effort against latency and resource use.

Google lists function calling, code execution, file search, Search grounding, URL context and structured output among its supported developer capabilities. Computer use is available in preview.

Gemini 3.7 Flash key specifications

Specification	Gemini 3.7 Flash
Status	Generally available
Model ID	gemini-3.7-flash
Inputs	Text, image, video, audio and PDF
Standard output	Text
Maximum input	1,048,576 tokens
Maximum output	65,536 tokens
Thinking levels	Low, medium and high
Function calling	Supported
Code execution	Supported
File search	Supported
Search grounding	Supported
Computer use	Preview
Main targets	Agentic tasks, coding, reasoning, multimodal understanding and knowledge work

These specifications describe the developer model. The consumer Gemini app abstracts many of these technical details from ordinary users.

Gemini remains a family of models

Gemini 3.7 Flash does not replace every other Gemini model.

Google continues to maintain several models aimed at different workloads. Current offerings include Gemini 3.6 Flash, Gemini 3.5 Flash, Gemini 3.5 Flash-Lite and specialist models for images, audio and other generative tasks. Gemini 3.1 Pro also remains listed as a preview model in Google’s API documentation.

Google describes Gemini 3.5 Flash-Lite as an efficiency-focused model for high-volume workloads, while 3.7 Flash targets more demanding reasoning, coding and agentic tasks.

That model diversity is important for developers. The largest or newest model is not automatically the best choice when speed, throughput and operational requirements matter.

The Gemini app does more than answer prompts

For consumers, Gemini increasingly behaves like a general AI workspace rather than a conventional question-and-answer chatbot.

Its current feature set includes tools for research, content creation, customised assistants and interaction with other services.

Deep Research

Deep Research can investigate a subject across multiple sources and produce a structured report.

Google Search acts as a default research source, while users can add files and, where supported, connect sources such as Gmail and Google Drive. NotebookLM notebooks can also be added as research material.

This makes the feature useful for comparing information across several sources, although generated reports still require human verification before important decisions.

Gems

Gems are customised versions of Gemini that follow reusable instructions.

A user could create separate Gems for recurring research, writing, business processes or other tasks rather than explaining the same requirements in every new conversation. Knowledge files can also be attached to a Gem.

Canvas and content creation

Gemini also includes Canvas for working on documents, prototypes and other generated material.

Google’s wider Gemini environment now supports features including image generation and editing, video creation, audio tools, quizzes, flashcards and Audio Overviews, although individual features can depend on account type, subscription and region.

Gemini Live

Gemini Live extends the assistant into conversational voice interactions. Google says users also employ Live with camera feeds and screen sharing for real-time assistance.

This can make Gemini more useful on a phone when describing an object, troubleshooting something visible on screen or talking through a task is easier than typing.

Gemini is becoming more agentic

One of Google’s biggest changes during 2026 has been moving Gemini from answering questions towards performing multi-step tasks.

Google has introduced Gemini Spark, an agent designed to continue working on assigned tasks in the cloud. Google describes Spark as capable of handling workflows even after the user closes a laptop or locks a phone.

Availability remains an important qualification. Google has been rolling Spark out progressively, and access can differ by country and subscription.

Google is also expanding Gemini’s Connected Apps system. In August 2026, it announced integrations spanning productivity, entertainment, local services and other third-party platforms.

An AI assistant with permission to act across several services can be more useful than a standalone chatbot. It also makes permission management and privacy settings considerably more important.

Gemini has specific relevance for South African users

The Gemini web app is officially supported in South Africa, and Google lists both Afrikaans and Zulu among the languages supported by the web service. The Gemini Android app is also officially available in South Africa.

Google expanded Personal Intelligence to South Africa in April 2026.

When enabled, Personal Intelligence can connect information from services such as Gmail, Google Photos, YouTube and Search. Gemini can then reason across information stored in those services to provide more personalised responses.

The feature is opt-in, and users decide which supported services to connect.

That local availability matters more than simply having access to a global AI website. Integration with existing Google accounts makes Gemini particularly relevant to South Africans already using Android, Gmail, Google Photos, Drive and other Google services.

Privacy settings deserve attention

Gemini can process substantial amounts of personal information, especially when users upload files or connect Google services.

Google’s current Gemini Privacy Hub says information provided to Gemini can include prompts, uploaded files, videos, photographs, screens and Gemini Live recordings or transcripts. Data from Connected Apps can also be processed when those services are enabled.

Google says a subset of collected data may be reviewed by trained human reviewers to improve and protect its services.

Users can control the Keep Activity setting. With Keep Activity disabled, Google says future chats are not used to train its AI models unless the user submits feedback. Those conversations can still be retained for up to 72 hours to provide the service and for safety purposes.

Where Keep Activity is enabled, Gemini activity is automatically deleted after 18 months by default. Google allows users to change that period or manually delete activity.

Businesses should therefore not treat a consumer AI account as an unrestricted destination for confidential corporate, customer or personal information. South African organisations also need to consider their own POPIA obligations before feeding personal information into any external AI service.

Who is Google Gemini for?

For ordinary users, Gemini can help with writing, brainstorming, researching, understanding documents and working across different media.

For students and researchers, Deep Research, file analysis and study tools can help organise information. AI-generated answers should still be checked against authoritative sources.

For professionals, Gemini becomes more useful when connected to existing Google workflows and customised with Gems or other productivity tools.

For developers, the Gemini API provides access to models such as Gemini 3.7 Flash for coding, document analysis, multimodal applications and agentic systems.

For businesses, Gemini Enterprise and Google Workspace integrations provide a route to deploy AI within managed organisational environments rather than relying entirely on personal consumer accounts.

Gemini still has important limitations

The strongest reason not to treat Gemini as an automatic replacement for human judgement is that generative AI can still produce incorrect information.

Google’s own Gemini 3.7 Flash model card identifies hallucinations as an ongoing limitation. The model can also experience occasional slowness or timeouts.

Model names and capabilities also change quickly. Google has already moved through several Gemini generations, and its API documentation contains formal deprecation schedules for older models. Developers should therefore avoid building a production system around an assumption that a particular model endpoint will remain unchanged indefinitely.

Consumers face a different issue: not every advertised Gemini capability is available to every account. Region, age, device, subscription and staged roll-outs can affect feature access.

Gemini is becoming an AI platform rather than a single chatbot

The most useful way to understand Gemini in 2026 is as Google’s broader AI platform.

At the consumer level, it is an assistant capable of research, voice conversations, content generation and interaction with connected services. Behind that interface sits a changing family of Gemini models.

For developers and organisations, Gemini extends further into APIs, coding, multimodal processing and AI agents.

Gemini 3.7 Flash demonstrates how quickly that model layer continues to change. For users, however, the more important question is increasingly not which model number sits underneath Gemini, but what information it can access, what actions it can perform and whether those permissions are appropriate for the task.