Google launches Gemini 3.6 Flash and Flash-Lite

Google has introduced Gemini 3.6 Flash, Gemini 3.5 Flash-Lite and a version specialized in cybersecurity, Gemini 3.5 Flash Cyber. All three models are designed to help AI agents work at lower cost, with less latency and fewer unnecessary steps.

The release is aimed mainly at businesses and developers running thousands of automated tasks: analyzing documents, writing code, searching for information or using a computer on the user’s behalf.

Gemini 3.6 Flash: more capability with fewer tokens

Gemini 3.6 Flash is the main model in this release. Google says it improves programming, information analysis and image and document handling compared with Gemini 3.5 Flash, while using fewer resources.

According to the Artificial Analysis index, it uses 17% fewer output tokens than its predecessor. Tokens are the units of text a model processes and generates, and their consumption directly affects price. In some tests, such as Datacurve’s DeepSWE, Google reports a reduction of up to 65%.

The model also needs fewer reasoning steps and fewer tool calls to complete multi-step tasks. Its price is $1.50 per million input tokens and $7.50 per million output tokens, below the price announced for Gemini 3.5 Flash.

In tests shared by Google, Gemini 3.6 Flash delivers better results than the previous version in several scenarios:

• DeepSWE: 49% versus 37%.
• MLE Bench, focused on machine learning research: 63.9% versus 49.7%.
• OSWorld-Verified, which evaluates computer use: 83.0% versus 78.4%.
• GDPval-AA v2, covering knowledge tasks: 1,421 versus 1,349.

These figures come from specific evaluations and do not guarantee the same result in every application. In practice, the model could be used to review contracts, interpret charts, draft reports or modify code with fewer execution loops.

Google is also adding new safeguards against misuse in areas such as cyberattacks and chemical, biological, radiological and nuclear threats. The company says it has tried to reduce refusals in legitimate use cases without making the model more vulnerable to evasion attempts.

Flash-Lite prioritizes speed and volume

Gemini 3.5 Flash-Lite is designed for fast, high-volume tasks, such as classifying thousands of documents, running internal searches or coordinating an agent’s subtasks.

Artificial Analysis measures a speed of 350 output tokens per second, making it the fastest model in the 3.5 series, according to Google. Its price is $0.30 per million input tokens and $2.50 per million output tokens.

Developers can choose from different reasoning levels. Lower levels reduce cost and latency for simple tasks, while higher levels support processes involving multiple steps or subagents.

Compared with Gemini 3.1 Flash-Lite, Google reports improvements in tests such as:

• Terminal-Bench 2.1: 54% versus 31%.
• GDM-MRCR v2, focused on long context: 72.2% versus 60.1%.
• GDPval-AA v2: 1,140 versus 642.

It also outperforms Gemini 3 Flash in some evaluations, including SWE-Bench Pro and OSWorld-Verified. That makes it an option for workloads where processing many requests quickly matters more than using the most capable model available.

A cybersecurity model with restricted access

Gemini 3.5 Flash Cyber is tuned to find, validate and fix software vulnerabilities. It integrates with CodeMender, Google’s code security agent, which coordinates multiple agents to create a combined report.

Google says the system achieves competitive performance at the highest level of the CyberGym benchmark. However, it will not be openly available. It will soon be offered only to governments and trusted partners through a limited pilot program.

The restriction reflects the dual-use nature of these tools. The same model that helps repair a vulnerability could make it easier to exploit if it falls into the wrong hands.

What changes for you

Both general-purpose models are already available to developers through the Gemini API, Google AI Studio and Android Studio. Gemini 3.6 Flash is also coming to Google Antigravity. Businesses can use them through the Gemini Enterprise Agent Platform, and Gemini 3.6 Flash is additionally available in the Gemini app.

Gemini 3.5 Flash-Lite is also beginning to roll out in Google Search. For you, the most visible effect will not necessarily be a new feature, but agents that respond faster, cost less to operate and can take on repetitive tasks at greater scale.

Google says Gemini 3.5 Pro is being tested with partners and that it is preparing for general availability. At the same time, the company has already begun its most ambitious training effort yet for Gemini 4.