Google launches three new Gemini models to enhance AI agent efficiency and speed.

Author: Google DeepMind

Compiled by Deep潮 TechFlow

Shenchao Overview: The core focus of Google’s three newly released models is addressing real commercial pain points for AI agents—cost and speed. The 3.6 Flash model uses 17% fewer tokens, is cheaper, and delivers better quality than its predecessor; the 3.5 Flash-Lite model achieves a speed of 350 tokens per second, making it the fastest in the 3.5 series to date; and the specialized cybersecurity model, Flash Cyber, targets the high-demand market of code vulnerability remediation. This is not a benchmarking contest—it’s about paving the way for large-scale deployment of AI agents.

Google DeepMind today released three new Gemini models: 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber. These models are designed for the efficiency, latency, and reliability required to build AI agents at scale.

3.6 Flash has been improved based directly on developer feedback for 3.5 Flash. It not only advances further in coding and knowledge work but also significantly improves token efficiency. According to the Artificial Analysis Index, 3.6 Flash reduces output token consumption by 17% compared to 3.5 Flash. In certain benchmarks, such as Datacurve’s DeepSWE, the reduction reaches up to 65%. It also requires fewer reasoning steps and tool calls to complete multi-step workflows.

This efficiency gain is accompanied by lower pricing: priced at $1.5 per million input tokens and $7.5 per million output tokens, 3.6 Flash reduces the total cost per Agent task, making it more economical to build and run Agents.

Even more efficient, 3.6 Flash outperforms 3.5 Flash across various use cases:

The code editing is more precise, reducing unnecessary modifications and execution loops, achieving 49% on DeepSWE (3.5 Flash: 37%) and significantly improving to 63.9% on the machine learning research benchmark MLE Bench (3.5 Flash: 49.7%).

Improved computer operation skills, OSWorld-Verified score of 83.0% (3.5 Flash at 78.4%). Computer operations are now available as an integrated client tool via Gemini API and Gemini Enterprise.

It performs better on knowledge-intensive tasks, such as achieving a GDPval-AA v2 benchmark score of 1421 (compared to 1349 for 3.5 Flash). Clients like Hebbia and Harvey have found it particularly strong in multimodal tasks including document parsing, chart and data analysis, and report drafting.

Customer feedback indicates that 3.6 Flash represents an advancement in both cost and quality, balancing token efficiency, accuracy, and speed in complex workflows and knowledge-intensive tasks.

3.6 Flash is equipped with enhanced state-of-the-art security measures covering chemical, biological, radiological, nuclear (CBRN), and cyber attack abuse domains. These protections significantly increase the model’s resistance to jailbreak attacks, while also being trained to minimize refusal of beneficial uses.

Gemini 3.5 Flash-Lite: Built for scaling agent workflows

3.5 Flash-Lite is designed for low-latency tasks and high-throughput development scenarios, such as agent search and document processing.

3.5 Flash-Lite is the fastest model in the 3.5 series, achieving a speed of 350 output tokens per second according to Artificial Analysis. Priced at $0.30 per million input tokens and $2.50 per million output tokens, it offers significantly better quality than 3.1 Flash-Lite, delivering exceptional value for developers and customers running high-volume production workloads.

3.5 Flash-Lite enables efficient scaling of the Agent system, significantly outperforming 3.1 Flash-Lite across all reasoning levels. Developers can configure the model based on workload: use minimal and low reasoning levels for bulk tasks to prioritize low latency and cost-effective execution, or enable higher reasoning levels to handle multi-step sub-Agent workloads. The model now also includes computer operation as a built-in tool, reliably supporting Agent tasks across interfaces.

It shows significant improvements in coding and agent tasks, such as a score of 54% on Terminal-Bench 2.1 (vs. 31% for 3.1 Flash-Lite), long-context performance like 72.2% on GDM-MRCR v2 (vs. 60.1% for 3.1 Flash-Lite), and real-world task execution with a score of 1140 on GDPval-AA v2 (vs. 642 for 3.1 Flash-Lite).

In fact, in many agent and coding benchmarks, 3.5 Flash-Lite outperforms 3 Flash, including on SWE-Bench Pro (54.2% vs. 49.6%) and OSWorld-Verified (74.0% vs. 65.1%), making it a faster and more powerful choice for 2.5 and 3 Flash workloads.

Early customers highlighted the unique combination of speed, intelligence, and cost efficiency of the 3.5 Flash-Lite in scaling agent workflows and data processing tasks.

Gemini 3.5 Flash Cyber in CodeMender: Efficiently detect and fix vulnerabilities

AI models are now discovering security vulnerabilities faster than current systems can patch them. Addressing this growing threat requires a software security approach that is both efficient and robust.

Flash's performance and efficiency make it an ideal foundation for large-scale detection, validation, and patching of code security issues. Gemini 3.5 Flash Cyber is built on 3.5 Flash, fine-tuned specifically to identify and remediate cybersecurity vulnerabilities, with a cost per token lower than that of larger models.

In CodeMender, multiple 3.5 Flash Cyber Agents work together to generate a single comprehensive report, achieving state-of-the-art performance on the popular benchmark CyberGym.

Given the dual-use nature of this technology, we have adopted a cautious deployment approach. The model will soon be made available exclusively to government agencies and trusted partners through CodeMender via a restricted-access pilot program. This will enable frontline defenders to detect and remediate critical vulnerabilities before they are exploited, while mitigating broader risks of misuse.

3.6 Flash and 3.5 Flash-Lite are available starting today:

Developers can access it via the Gemini API in Google AI Studio and Android Studio. 3.6 Flash is also available in Google Antigravity.

Enterprise users can access 3.6 Flash on the Gemini Enterprise Agent platform. 3.6 Flash is also available in the Gemini Enterprise app.

Available to everyone through the Gemini app. 3.5 Flash-Lite is also rolling out on Google Search.

Please provide feedback to help improve future Gemini models—we look forward to releasing 3.5 Pro soon.