Gemini 3.6 and 3.5: Revolutionizing AI Efficiency with Flash and Cyber

⚡

Key Takeaways

1The new Gemini 3.6 Flash and 3.5 Flash-Lite models promise increased efficiency for AI agents.

2Gemini 3.5 Flash Cyber specializes in cybersecurity, optimizing the detection and remediation of vulnerabilities.

3The models are available for developers and businesses through Google AI Studio and Gemini Enterprise.

💡Why it matters — These innovations enhance developers' ability to create faster and more secure AI solutions, which is crucial for technological advancement.

A New Era for Gemini Models

The latest models in the Gemini series, namely Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber, are designed to meet the growing demands for efficiency, latency, and reliability of large-scale artificial intelligence agents. For developers and clients relying on these agents in production, improving token efficiency, reducing latency, and ensuring reliable performance are imperatives. The Flash series of Gemini aims to provide an optimal balance between efficiency and quality, enabling a smooth evolution of agent workflows.

New Models in Detail

3.6 Flash: This flagship model stands out for its superior performance in coding, knowledge work, and multimodal capabilities. According to the Artificial Analysis Index, it manages to reduce output token usage by 17% compared to its predecessor, the 3.5 Flash. In certain benchmarks, such as Datacurve's DeepSWE, this reduction reaches up to 65%, while maintaining a lower cost per output token.
3.5 Flash-Lite: This model is the fastest and most cost-effective in the 3.5 series, capable of producing 350 output tokens per second, according to the Artificial Analysis Index. It outperforms previous versions of Flash-Lite in agent workflows, providing a quick and economical solution.
3.5 Flash Cyber: Integrated into CodeMender, this model focuses on cybersecurity. It combines a specialized model with an agent infrastructure to deliver top-notch performance in detecting and correcting security vulnerabilities in code.

Performance and Innovations of Gemini 3.6 Flash

Gemini 3.6 Flash was developed in response to user feedback from 3.5 Flash, featuring notable improvements in coding and knowledge work, as well as increased token efficiency. On the Artificial Analysis Index, it consumes 17% fewer output tokens than the previous model. Additionally, it requires fewer reasoning steps and tool calls to accomplish complex tasks.

This efficiency translates into reduced costs, with a price of $1.50 for one million input tokens and $7.50 for one million output tokens, making the creation and execution of agents more affordable.

Performance Improvements

Gemini 3.6 Flash outperforms its predecessor in several application areas:

Increased Accuracy: Fewer undesirable code modifications and reduced execution loops, as shown in the DeepSWE benchmark (49% vs. 37%).
Enhanced ML Search: Significant results in MLE Bench (63.9% vs. 49.7%).
Computational Usability Capabilities: Demonstrated improvements by OSWorld-Verified (83.0% vs. 78.4%).

In the realm of knowledge work, 3.6 Flash excels, as indicated by benchmarks such as GDPval-AA v2 (1421 vs. 1349). Clients like Hebbia and Harvey have noted its exceptional performance in document analysis, graphs, and data, as well as in report writing.

3.5 Flash-Lite: Optimizing Workflows

Gemini 3.5 Flash-Lite is designed for tasks requiring low latency and high throughput, essential for developers in fields such as agent research and document processing. This model, the fastest in the 3.5 series, achieves a speed of 350 output tokens per second. With a cost of $0.30 for one million input tokens and $2.50 for one million output tokens, it offers excellent value for high-throughput production environments.

3.5 Flash Cyber: Enhanced Security

AI models like Gemini 3.5 Flash Cyber are now capable of detecting security vulnerabilities faster than current systems can correct them. Thanks to its performance and efficiency, Flash is ideal for detecting, validating, and correcting security issues in code at scale.

Gemini 3.5 Flash Cyber, based on 3.5 Flash, is optimized for cybersecurity, offering a lower cost per token than larger models. In CodeMender, which utilizes multiple 3.5 Flash Cyber agents to produce a combined report, this model achieves competitive performance on the CyberGym benchmark.

Availability of New Models

The 3.6 Flash and 3.5 Flash-Lite models are now available:

For developers, via the Gemini API in Google AI Studio and Android Studio. 3.6 Flash is also accessible in Google Antigravity.
For enterprises, on the Gemini Enterprise agent platform. 3.6 Flash is also available in the Gemini Enterprise application.
For the general public, via the Gemini application. 3.5 Flash-Lite is also being deployed in Google Search.

We encourage users to share their feedback to improve future Gemini models, and we are preparing for the upcoming launch of 3.5 Pro.