Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber release notes

Google DeepMind has introduced Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber to provide the efficiency, latency, and reliability required to scale production AI agents. These models are designed to optimize the balance between quality and cost, specifically targeting agentic workflows, coding, and cybersecurity.

Gemini 3.6 Flash: Enhanced Efficiency and Performance

Gemini 3.6 Flash is designed as a workhorse model that improves upon Gemini 3.5 Flash by reducing output token usage and improving performance in coding and knowledge work.

Token Efficiency and Cost

Gemini 3.6 Flash reduces output token usage by 17% compared to 3.5 Flash according to the Artificial Analysis Index. In specific benchmarks like DeepSWE by Datacurve, output token reduction is observed up to 65%. The model is priced at $1.50 per 1 million input tokens and $7.50 per 1 million output tokens, making it more cost-effective than its predecessor.

Performance Benchmarks

3.6 Flash demonstrates gains across several key use cases:

• Coding and ML Research: It achieves 49% on DeepSWE (compared to 37% for 3.5 Flash) and 63.9% on MLE Bench (compared to 49.7% for 3.5 Flash).
• Computer Use: It reaches 83.0% on OSWorld-Verified (compared to 78.4% for 3.5 Flash). Computer use is now integrated as a built-in client-side tool via Gemini Enterprise and the Gemini API.
• Knowledge Work: It scores 1421 on GDPval-AA v2, outperforming 3.5 Flash's 1349.

Safety Framework

3.6 Flash includes enhanced Frontier Safety safeguards specifically targeting cyber offense and Chemical, Biological, Radiological, and Nuclear (CBRN) misuses to increase resistance to jailbreaks while minimizing refusals for beneficial use cases.

Gemini 3.5 Flash-Lite: High-Throughput Agentic Scaling

Gemini 3.5 Flash-Lite is the fastest model in the 3.5 series, optimized for low-latency tasks and high-throughput workflows such as document processing and agentic search.

Speed and Cost

According to the Artificial Analysis Index, 3.5 Flash-Lite delivers 350 output tokens per second. It is priced at $0.30 per 1 million input tokens and $2.50 per 1 million output tokens.

Agentic Capabilities

3.5 Flash-Lite supports configurable thinking levels, allowing developers to prioritize either low-latency execution for high-volume tasks (minimal/low thinking levels) or higher thinking levels for multi-step subagent workloads. It also features computer use as a built-in tool.

Performance Improvements

3.5 Flash-Lite outperforms prior generations and some larger models in specific agentic and coding tasks:

• Coding: 54% on Terminal-Bench 2.1 (up from 31% for 3.1 Flash-Lite).
• Long Context: 72.2% on GDM-MRCR v2 (up from 60.1% for 3.1 Flash-Lite).
• Real-world Execution: 1140 on GDPval-AA v2 (up from 642 for 3.1 Flash-Lite).
• Comparative Performance: It outperforms 3 Flash on SWE-Bench Pro (54.2% vs 49.6%) and OSWorld-Verified (74.0% vs 65.1%).

Gemini 3.5 Flash Cyber and CodeMender

Gemini 3.5 Flash Cyber is a specialized model fine-tuned for detecting, validating, and patching cybersecurity vulnerabilities.

Integration with CodeMender

The model is integrated into CodeMender, a code security agent that uses multiple 3.5 Flash Cyber agents to produce combined security reports. This combination achieves competitive frontier performance on the CyberGym benchmark.

Deployment and Access

Due to the dual-use nature of the technology, 3.5 Flash Cyber is not broadly available. It is exclusively available to governments and trusted partners through a limited-access pilot program via CodeMender.

Availability and Future Roadmap

Gemini 3.6 Flash and 3.5 Flash-Lite are available immediately via the Gemini API (Google AI Studio and Android Studio), Google Antigravity, the Gemini Enterprise Agent Platform, the Gemini Enterprise app, and the Gemini app. 3.5 Flash-Lite is also rolling out in Google Search.

Google DeepMind has announced that Gemini 3.5 Pro is currently testing with partners and will be broadly available soon. Additionally, pre-training has begun for the next-generation Gemini 4.