Google Introduces Three New Gemini AI Models for Faster and More Affordable AI Agents

Google Introduces Three New Gemini AI Models for Faster and More Affordable AI Agents

Google has expanded its artificial intelligence lineup with three new Gemini models designed for coding, high-volume automation and cybersecurity.

The newly introduced models are:

• Gemini 3.6 Flash
• Gemini 3.5 Flash-Lite
• Gemini 3.5 Flash Cyber

Gemini 3.6 Flash is Google's new general-purpose workhorse model for coding, knowledge work and multimodal tasks. Gemini 3.5 Flash-Lite is designed for fast and affordable automation at scale, while Gemini 3.5 Flash Cyber focuses on detecting and repairing software vulnerabilities.

The announcement reflects Google's growing focus on agentic AI—systems capable of planning tasks, using software tools and completing multi-step workflows with less human involvement.

However, the three models do not have the same availability. Gemini 3.6 Flash and Gemini 3.5 Flash-Lite are generally available, while Gemini 3.5 Flash Cyber will initially be restricted to governments and selected trusted partners through a limited pilot program.

Google's Three New Gemini Models at a Glance

The prices shown are Google's published Gemini API rates and may differ across platforms, regions or enterprise agreements.

Gemini 3.6 Flash: Google's New Workhorse Model

Gemini 3.6 Flash is the most broadly capable model in the announcement.

Google developed it as an improved version of Gemini 3.5 Flash, with better performance in coding, knowledge work, multimodal understanding and computer-based tasks.

The model accepts several types of input, including:

• Text
• Images
• Audio
• Video
• PDF documents

It produces text output and supports capabilities such as code execution, function calling, structured responses, Google Search grounding, file search, URL context and computer use.

Gemini 3.6 Flash also supports an input context of more than one million tokens. This enables developers to process large documents, extensive codebases, long conversations and collections of files within a single workflow.

Improved Token Efficiency

One of the most important improvements in Gemini 3.6 Flash is token efficiency.

Tokens are the small units of text processed and generated by AI models. Using fewer tokens can lower costs and reduce the time required to complete a task.

Google says Gemini 3.6 Flash uses approximately 17% fewer output tokens than Gemini 3.5 Flash on the Artificial Analysis Index. The company also reports larger reductions in certain software-engineering evaluations.

According to Google, the model can complete multi-step assignments using:

• Fewer reasoning steps
• Fewer tool calls
• Shorter responses
• Fewer coding corrections
• Reduced execution loops

These improvements may be particularly valuable for businesses operating AI agents at scale. Even a small reduction in tokens per request can create meaningful savings when an application processes millions of requests.

Better Coding and Computer Use

Coding is one of the main areas Google has targeted with Gemini 3.6 Flash.

The model is designed to produce more precise code while making fewer unnecessary changes to surrounding files. This can make it more suitable for tasks such as:

• Generating application features
• Debugging existing code
• Migrating software projects
• Reviewing pull requests
• Refactoring codebases
• Creating automated developer agents
• Performing machine-learning engineering tasks

Google also reports improved computer-use performance. This means Gemini 3.6 Flash can support agents that interact with browsers, desktop environments and mobile interfaces.

Computer use is available as a built-in client-side tool through the Gemini API and Gemini Enterprise. Developers remain responsible for adding appropriate permissions, monitoring and confirmation steps before allowing an agent to perform sensitive actions.

Gemini 3.6 Flash Pricing

Google has priced Gemini 3.6 Flash at:

• $1.50 per one million input tokens
• $7.50 per one million output tokens

The input price is similar to Gemini 3.5 Flash, but the output price is lower. Combined with reduced token usage, Google says the new model should lower the overall cost of completing agentic tasks.

Gemini 3.6 Flash is likely to be most useful when a project needs a balance between intelligence, speed and operating cost without moving to a more expensive Pro-level model.

Gemini 3.5 Flash-Lite: Built for High-Volume Automation

Gemini 3.5 Flash-Lite is the fastest and least expensive model in Google's 3.5 Flash family.

It is designed for applications that need to process large numbers of relatively lightweight tasks with minimal delay.

Potential use cases include:

• Customer-support classification
• Document processing
• Receipt translation
• Product-data extraction
• Search agents
• Content tagging
• Summarization
• Data validation
• Information routing
• Subagents within larger AI systems

Google cites testing from Artificial Analysis that measured the model at approximately 350 output tokens per second. Actual speed will depend on the application, request, selected thinking level and available infrastructure.

Configurable Thinking Levels

Gemini 3.5 Flash-Lite allows developers to adjust how much reasoning the model uses.

Minimal or low thinking can be selected for simple, time-sensitive and high-volume requests. Higher thinking levels can be used when a subagent needs to complete a more complicated sequence of tasks.

This flexibility allows developers to choose between:

• Faster responses
• Lower API costs
• More detailed reasoning
• Better handling of multi-step tasks

For example, a SaaS application could use Gemini 3.6 Flash as its main coordinating agent while assigning repetitive tasks to several lower-cost Gemini 3.5 Flash-Lite subagents.

The main agent could plan the workflow, while the Flash-Lite models process documents, extract data or generate alternative results.

Gemini 3.5 Flash-Lite Pricing

Google has priced Gemini 3.5 Flash-Lite at:

• $0.30 per one million input tokens
• $2.50 per one million output tokens

This makes it considerably less expensive than Gemini 3.6 Flash.

It may therefore be the better choice for startups that process large numbers of requests but do not require the strongest available reasoning for every task.

The lowest-cost model is not automatically the best option. Developers should test output quality, latency and reliability using workloads that represent their actual applications.

Gemini 3.5 Flash Cyber: AI for Software Security

The most specialized model in the announcement is Gemini 3.5 Flash Cyber.

It is based on Gemini 3.5 Flash but has been fine-tuned to detect, validate and repair cybersecurity vulnerabilities.

Google is combining the model with CodeMender, its AI-powered code-security agent. CodeMender can coordinate multiple Gemini 3.5 Flash Cyber agents and combine their findings into a single security report.

Possible defensive uses include:

• Identifying vulnerable code
• Confirming whether a vulnerability is exploitable
• Suggesting security patches
• Reviewing software dependencies
• Detecting unsafe programming patterns
• Helping maintainers repair open-source projects
• Prioritizing critical security problems

Google says the model achieves competitive results on CyberGym, a benchmark used to evaluate cybersecurity capabilities.

Why Flash Cyber Will Have Restricted Access

Cybersecurity AI is considered dual-use technology.

The same capabilities that can help defenders discover and repair vulnerabilities could potentially be misused to find weaknesses for malicious attacks.

Because of this risk, Gemini 3.5 Flash Cyber will not initially be released through the public Gemini API. Google plans to provide it through CodeMender to government organizations and selected trusted partners as part of a limited-access pilot.

The company says this controlled rollout is intended to give defenders access to advanced vulnerability-detection tools while reducing the risk of misuse.

Broader availability may come later, but Google has not announced a public release date or standard API pricing.

Safety Improvements in Gemini 3.6 Flash

Google says Gemini 3.6 Flash includes stronger Frontier Safety protections.

These safeguards focus on high-risk areas such as:

• Chemical threats
• Biological threats
• Radiological and nuclear risks
• Offensive cybersecurity misuse
• Attempts to bypass model restrictions

The company says it has worked to make the model more resistant to jailbreaks while reducing unnecessary refusals for legitimate requests.

These protections are important as Gemini models become more capable of using computers, writing code and performing actions across external tools.

Where Are the New Models Available?

Gemini 3.6 Flash and Gemini 3.5 Flash-Lite are available through several Google platforms.

For developers

The models can be accessed through:

• Gemini API
• Google AI Studio
• Android Studio
• Google Antigravity for Gemini 3.6 Flash

For businesses

Enterprise customers can use the models through:

• Gemini Enterprise Agent Platform
• Gemini Enterprise application for Gemini 3.6 Flash

For general users

Both models are available through the Gemini application. Gemini 3.5 Flash-Lite is also being introduced gradually into Google Search.

Availability, features and usage limits may vary depending on the user's location, account and subscription.

Which Gemini Model Should Developers Choose?

The best model depends on the type of application being developed.

Choose Gemini 3.6 Flash when:

• You need advanced coding capabilities.
• Your agent must complete complex workflows.
• You are analyzing documents, charts or mixed media.
• You need computer-use capabilities.
• Quality is more important than obtaining the lowest possible cost.

Choose Gemini 3.5 Flash-Lite when:

• You need very fast responses.
• Your application processes a high volume of requests.
• Cost efficiency is the main concern.
• You are building subagents for repetitive tasks.
• The workload involves extraction, translation or summarization.

Consider Gemini 3.5 Flash Cyber when:

• Your organization is eligible for Google's restricted pilot.
• You need AI-assisted vulnerability detection and patching.
• Your use case is defensive cybersecurity.
• You can apply strong oversight and security controls.

What About Gemini 3.5 Pro?

Gemini 3.5 Pro was not included in the general release.

Google says the model is currently being tested with partners and will become broadly available once it is ready. The company has not provided a confirmed public launch date.

Google also revealed that work has begun on Gemini 4, including what it describes as its most ambitious model pre-training run so far.

For now, Gemini 3.6 Flash appears to be Google's main option for developers who need advanced intelligence without waiting for Gemini 3.5 Pro.

What This Launch Means for the AI Industry

Google's latest release is not focused on producing one model for every possible task.

Instead, the company is building a family of specialized models for different levels of intelligence, speed, cost and risk.

Gemini 3.6 Flash provides the strongest general-purpose performance of the three. Gemini 3.5 Flash-Lite handles inexpensive, high-volume tasks, while Gemini 3.5 Flash Cyber targets a highly specialized and sensitive industry.

This approach could help developers build AI systems in which a capable main agent coordinates several faster and less expensive subagents.

Final Thoughts

Google's three new Gemini models demonstrate how the AI industry is moving from basic chatbots toward systems that can plan, use tools and complete complicated workflows.

Gemini 3.6 Flash delivers improvements in coding, multimodal understanding, computer use and token efficiency. Gemini 3.5 Flash-Lite gives developers a faster and more affordable option for automation at scale.

Gemini 3.5 Flash Cyber extends the Flash family into cybersecurity, although its rollout will remain restricted because of the risks associated with advanced vulnerability-discovery tools.

For developers and businesses, the announcement provides more choice when balancing performance, speed, safety and cost. It also shows that Google's next major competition in AI may centre less on chatbot answers and more on the agents that can reliably perform real work.