Gemini 3.6 Flash: Faster, Cheaper AI Agents

Google has unveiled its latest suite of Gemini models, headlined by Gemini 3.6 Flash. This release focuses on delivering the efficiency, low latency, and reliability necessary for building scalable AI agents, addressing critical developer demands for production-grade applications.

The new Flash series aims to strike an optimal balance between performance and cost. It introduces several key models designed for distinct use cases, pushing the boundaries of what developers can achieve with Google's AI.

Gemini 3.6 Flash: The Efficiency Workhorse

Gemini 3.6 Flash is positioned as Google's new workhorse model, offering significant improvements in coding, knowledge work, and multimodal performance. It boasts a 17% reduction in output token usage compared to 3.5 Flash, as measured by the Artificial Analysis Index, with some benchmarks like DeepSWE by Datacurve showing up to a 65% reduction. This efficiency translates to a lower cost per output token.

The model also demonstrates enhanced performance across various metrics. It delivers higher precision in code edits (49% vs. 37% in DeepSWE), improved ML Research capabilities (63.9% vs. 49.7% in MLE Bench), and better computer use capabilities (83.0% vs. 78.4% in OSWorld-Verified). For knowledge work, it outperforms 3.5 Flash, scoring 1421 vs. 1349 in GDPval-AA v2. Customers like Hebbia and Harvey have reported its strength in multimodal tasks such as document parsing and data analysis.

Security remains a priority, with 3.6 Flash incorporating enhanced Frontier Safety safeguards against Chemical, Biological, Radiological, and Nuclear (CBRN) and cyber offense misuses. These measures significantly bolster the model's resistance to jailbreaks while minimizing refusals for beneficial applications.

Gemini 3.5 Flash-Lite: Speed and Cost-Effectiveness

For scenarios demanding extreme speed and cost efficiency, Google introduces Gemini 3.5 Flash-Lite. This model is engineered for low-latency and high-throughput tasks, such as agentic search and document processing, executing at 350 output tokens per second according to Artificial Analysis.

Priced at $0.3/1M input tokens and $2.5/1M output tokens, 3.5 Flash-Lite offers a compelling price-to-performance ratio. It significantly surpasses earlier Flash-Lite generations in agentic workflows and even outperforms 3 Flash in several agentic and coding evaluations, including SWE-Bench Pro (54.2% vs. 49.6%) and OSWorld-Verified (74.0% vs. 65.1%).

Gemini 3.5 Flash Cyber in CodeMender: Cybersecurity Focus

Addressing the growing challenge of cybersecurity, Google is launching Gemini 3.5 Flash Cyber, a specialized model fine-tuned for finding and fixing vulnerabilities. Integrated into CodeMender, Google's code security agent, this model achieves competitive performance on benchmarks like CyberGym.

Due to its dual-use nature, 3.5 Flash Cyber will be available exclusively to governments and trusted partners through a limited-access pilot program. This strategic deployment aims to empower frontline defenders in proactive vulnerability management.

Availability and Future Outlook

Gemini 3.6 Flash and 3.5 Flash-Lite are available immediately to developers via the Gemini API, Google AI Studio, and Android Studio. Enterprise users can access 3.6 Flash through the Gemini Enterprise Agent Platform and the Gemini Enterprise app, while 3.5 Flash-Lite is rolling out in Google Search.

Google also confirmed that Gemini 3.5 Pro is currently undergoing partner testing, with a broader release planned soon. The company has already initiated its most ambitious pre-training run yet for Gemini 4, signaling continued advancements in AI model development.