Google rolls out Gemini 3.6 Flash and a limited cybersecurity model

Google has introduced a new set of Gemini AI models aimed at faster, cheaper developer use, including its first Gemini version built for cybersecurity work. The release matters for developers and businesses watching AI costs, because Google says the new models use fewer tokens while improving some coding and computer-use benchmarks.

The company announced Gemini 3.6 Flash, Gemini 3.5 Flash Lite and Gemini 3.5 Flash Cyber in a company blog post. Google did not include Gemini 3.5 Pro, the higher-end model it had previously said at I/O would arrive in June, according to Ars Technica.

Gemini 3.6 Flash replaces Gemini 3.5 Flash, which Google had highlighted at I/O in May. Google says the new Flash model is more capable, stronger at coding and better suited for multimodal work, with changes shaped by feedback from users of the prior version.

Google said Gemini 3.6 Flash scores 49 percent on the DeepSWE coding test, up from 37 percent for Gemini 3.5 Flash. The company also said the new model reaches 83 percent on OSWorld, a computer-use benchmark, compared with 78.4 percent for the previous Flash model.

The company is also emphasizing cost controls. Google says Gemini 3.6 Flash uses about 17 percent fewer tokens and should finish agent-style tasks with fewer steps and less token usage.

Google set Gemini 3.6 Flash API pricing at $1.50 per 1 million input tokens and $7.50 per 1 million output tokens. Gemini 3.5 Flash had the same input price and a $9 output price, according to the figures cited by Google.

Flash Lite targets high-volume AI use

Google also released Gemini 3.5 Flash Lite, which it describes as its most efficient modern AI model. The company says Flash Lite can generate 350 tokens per second and is intended for scaling agent-based systems at lower cost.

Google priced Gemini 3.5 Flash Lite at $0.30 per 1 million input tokens and $2.50 per 1 million output tokens. Ars Technica noted that this is higher than the previous Gemini 3.1 Flash Lite pricing of $0.25 for input and $1.50 for output.

Google said Gemini 3.6 Flash is rolling out through the API and will replace Gemini 3.5 Flash in the Gemini app. The company also said Gemini 3.5 Flash Lite is available to developers and in the Gemini app, and that the model will appear frequently in Google Search, where its speed may suit AI Overviews.

Cybersecurity model gets a restricted launch

Gemini 3.5 Flash Cyber is Google’s first large language model tuned specifically for cybersecurity, according to the company. Google says the model is close to Anthropic’s Claude Mythos in finding and fixing cybersecurity issues while keeping the efficiency profile of a Flash model.

Google acknowledged that cybersecurity AI can be used both to defend systems and to find weaknesses for harmful purposes. The company said Gemini 3.5 Flash Cyber will launch first as a limited pilot inside Google DeepMind’s CodeMender agent for trusted partners and governments.

The delayed Gemini 3.5 Pro remains unresolved. Google said the model is being tested with unnamed partners and will be released when it is ready, while Ars Technica reported that earlier accounts said Google had delayed it because it was not matching rival models in coding.

Google also said it has begun pre-training Gemini 4, describing that effort as more ambitious than prior work. The company did not give a release timeline for Gemini 4 or say whether more Gemini 3.x models will arrive before then.

This story draws on original reporting from Ars Technica.