In an official blog post published by Tulsee Doshi, Senior Director of Product Management on behalf of the Gemini team, Google has introduced its latest artificial intelligence models: Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber. These models were developed to provide developers and customers building production-grade AI agents with higher token efficiency, lower latency, and more reliable performance. Google’s Flash model series is designed to strike an optimal balance between efficiency and quality to enable the scaling of agentic workflows.
Gemini 3.6 Flash: Improved Efficiency and Higher Quality
The Gemini 3.6 Flash model was built directly based on feedback received from developers and customers who utilized the 3.5 Flash model. The new model introduces a step up in coding capabilities, knowledge tasks, and multimodal performance, while delivering a meaningful improvement in token efficiency.
According to data from the Artificial Analysis Index, Gemini 3.6 Flash consumes 17% fewer output tokens compared to 3.5 Flash. In specific benchmarks, such as Datacurve's DeepSWE index, output token consumption dropped by up to 65%. Additionally, the model requires fewer reasoning steps and fewer tool calls to complete multi-step workflows.
This enhanced token efficiency is accompanied by a reduction in costs compared to the 3.5 Flash model: Gemini 3.6 Flash is priced at $1.50 per million input tokens and $7.50 per million output tokens. This pricing reduces the overall cost of agentic tasks, making it highly cost-effective to build and run these agents.
Alongside the efficiency gains, the model demonstrates performance increases across a wide range of benchmarks:
- On the DeepSWE index, the model delivers higher precision with fewer unwanted code edits and reduced execution loops (achieving a 49% success rate compared to 37% in the previous model).
- In machine learning research, as measured by the MLE Bench index, performance improved to 63.9% compared to 49.7%.
- Computer use capabilities rose to 83.0% on the OSWorld-Verified index, compared to 78.4% previously. This feature is now available as a built-in client-side tool via the Gemini API and Gemini Enterprise.
- On knowledge tasks, as demonstrated by benchmarks like GDPval-AA v2, the model achieved a score of 1421 compared to 1349.
Google enterprise customers, such as Hebbia and Harvey, reported that the new model demonstrates enhanced capabilities in multimodal tasks, including document parsing, chart and data analysis, and report drafting. Furthermore, the model assists in analyzing financial data and transcripts more efficiently and accurately using Managed Agents on AIS, and enables running code migrations with lower latency and higher quality on AGY.
Built-In Safety Safeguards
Gemini 3.6 Flash is shipping with enhanced Frontier Safety safeguards designed to prevent misuse in the domains of offensive cyber capabilities as well as chemical, biological, radiological, and nuclear (CBRN) materials. These safeguards make the model substantially more resistant to jailbreaks, while the model has been trained to minimize refusals for beneficial and legitimate use cases.
Gemini 3.5 Flash-Lite: Speed and Savings for Large-Scale Workloads
In addition to the 3.6 model, Google is launching Gemini 3.5 Flash-Lite, designed for tasks requiring exceptionally low latency or high-throughput developer workloads, such as agentic search and document processing. This model is the fastest in the 3.5 series, running at 350 output tokens per second according to Artificial Analysis data.
The pricing for this model is set at $0.3 per million input tokens and $2.5 per million output tokens. Gemini 3.5 Flash-Lite presents a significant improvement in quality compared to Gemini 3.1 Flash-Lite, enabling cost-effective agent deployment. Developers can configure the model to prioritize low-latency, low-cost execution for high-volume tasks using minimal and low thinking levels, or engage higher thinking levels to process complex multi-step subagent workloads. The model also now includes computer use as a built-in tool to reliably support these agentic tasks.
In benchmark tests, Gemini 3.5 Flash-Lite demonstrates a significant step up in coding and agentic tasks:
- On the Terminal-Bench 2.1 index, it achieved 54% compared to 31% for the previous model.
- In long context scenarios on the GDM-MRCR v2 index, it recorded 72.2% compared to 60.1%.
- In real-world task execution on the GDPval-AA v2 index, it rose to 1140 compared to 642.
On many coding and agentic benchmarks, the new model even outperforms Gemini 3 Flash, including on SWE-Bench Pro (54.2% vs. 49.6%) and OSWorld-Verified (74.0% vs. 65.1%). Early customers highlight the model's unique combination of speed and cost efficiency for tasks such as product feature extraction from massive e-commerce datasets, receipt translation and summarization, or quickly generating web designs and games.
Gemini 3.5 Flash Cyber: Finding and Patching Security Vulnerabilities in CodeMender
AI systems are currently capable of discovering security vulnerabilities faster than existing systems can patch them. To address this growing threat, Google developed Gemini 3.5 Flash Cyber. The model is built on top of the 3.5 Flash foundation and has been specifically fine-tuned to locate and patch cybersecurity vulnerabilities at a lower cost per token than larger models.
The model operates within the CodeMender system, which orchestrates multiple Gemini 3.5 Flash Cyber agents working together to produce a single unified security report. The system delivers competitive performance on the popular CyberGym benchmark. Due to the dual-use nature of this technology, Google is taking a controlled approach to its deployment: the model will soon be exclusively available to governments and trusted partners as part of a limited-access pilot program, aiming to support frontline defenders in finding and fixing critical vulnerabilities before they can be exploited.
Future Roadmap: Gemini 3.5 Pro and Gemini 4
Beyond these current releases, Google notes that Gemini 3.5 Pro is currently undergoing testing with selected partners, and the company plans to make it broadly available as soon as it is ready. In parallel, development teams have started their most ambitious pre-training run yet for the next-generation model, Gemini 4.
Availability and Access to the New Models
The Gemini 3.6 Flash and Gemini 3.5 Flash-Lite models are available starting today across several channels:
- For developers: Via the Gemini API in Google AI Studio and Android Studio. Gemini 3.6 Flash is also available in Google Antigravity.
- For enterprises: Within the Gemini Enterprise Agent Platform. Gemini 3.6 Flash is also available in the Gemini Enterprise app.
- For everyone: Via the Gemini app. Additionally, Gemini 3.5 Flash-Lite is being gradually rolled out within Google Search.