Google has launched three new AI models under the Gemini lineup: Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber.
Gemini 3.6 Flash and Gemini 3.5 Flash-Lite are available immediately through the Gemini API in Google AI Studio and Android Studio, Gemini Enterprise Agent Platform, and the Gemini app. Google said it will also begin rolling out Flash-Lite in Google Search.
The company said Gemini 3.6 Flash replaces Gemini 3.5 Flash as its primary workhorse model, offering lower token usage, lower inference costs, and improved performance across coding, knowledge work and multimodal tasks.
“Our newest Gemini models deliver the efficiency, latency, and reliability to build AI agents at scale,” Tulsee Doshi, Senior Director of Product Management at Google, wrote in a blog post.
Gemini 3.5 Flash-Lite targets high-throughput workloads with lower latency, while Gemini 3.5 Flash Cyber is a specialised model for cybersecurity applications. Modelled as a rival to Anthropic’s Claude Mythos, it will initially be available only to governments and selected partners through Google’s CodeMender AI code security agent that can scan and fix software vulnerabilities.
“AI models have become capable of finding security vulnerabilities faster than current systems can fix them,” Google said. “Tackling this growing threat requires an approach to securing software that is highly capable and efficient.”
Google also disclosed that Gemini 4 has entered pre-training, calling it its “most ambitious pre-training run yet”, while Gemini 3.5 Pro remains in testing with partners ahead of a broader release.
Gemini 3.6 Flash is priced at $1.50 per million input tokens and $7.50 per million output tokens, lower than Gemini 3.5 Flash. Google said the model uses 17% fewer output tokens than its predecessor on the Artificial Analysis Index, which offers a holistic measure of AI capabilities, and requires fewer reasoning steps and tool calls for multi-step workflows.
According to the company, Gemini 3.6 Flash improved its score on software engineering benchmark DeepSWE to 49% from 37%, OpenAI’s MLE Bench to 63.9% from 49.7%, edge applications-focused OSWorld-Verified to 83.0% from 78.4%, and GDPval-AA v2 to 1,421 from 1,349 on real-world evaluations.
Google added that customers including Harvey and Hebbia have used the model for document parsing, chart analysis, and report drafting.
Google said the model also includes updated safeguards against chemical, biological, radiological, nuclear, and cyber-offensive misuse while reducing refusals for legitimate use cases.
For cost-sensitive workloads, Gemini 3.5 Flash-Lite costs $0.30 per million input tokens and $2.50 per million output tokens. The company said the model generates 350 output tokens per second, according to Artificial Analysis, and is intended for applications such as document processing, agentic search and large-scale automation.
Google said Flash-Lite outperformed Gemini 3 Flash on several benchmarks, including SWE-Bench Pro (54.2% versus 49.6%) and OSWorld-Verified (74.0% versus 65.1%), while also improving over Gemini 3.1 Flash-Lite on coding, long-context reasoning, and real-world task execution.
ALSO READ: Alteryx Inspire 2026: Three Questions Every Data Leader Should Take to Orlando
