Google rolls out Gemini 3.6 Flash with lower token use

Google is pushing Gemini back into the model race with Gemini 3.6 Flash, a new version of its flagship model that the company describes as its “workhorse” option for balancing quality and efficiency. The headline claim is cost-related: Google says Gemini 3.6 Flash can reduce output tokens by as much as 65% compared with 3.5 Flash in some uses, and uses 17% fewer output tokens overall than the previous model.

That matters as customers pay closer attention to the cost of running models. But on performance, the picture is less impressive. According to the source article, Gemini 3.6 Flash trails Anthropic’s Claude Sonnet 5 and OpenAI’s GPT-5.6 on most major benchmark tests, and also falls behind Grok 4.5 in areas such as agentic coding. Pricing does not appear to give Google much of an edge either, with the model described as roughly in line with Grok 4.5 and GPT-5.6.

Other Gemini 3.5 models announced

Google also introduced two additional models built on the previous generation:

• Gemini 3.5 Flash-Lite, which Google calls its “fastest, most cost-effective” model
• Gemini 3.5 Flash Cyber, a cybersecurity-focused model

Google says 3.5 Flash-Lite improves efficiency while building on the prior version’s performance. 3.5 Flash Cyber is being kept on tighter controls: the company says it will be “exclusively available to governments and trusted partners” through its CodeMender AI security agent as part of a pilot program aimed at detecting and patching security vulnerabilities.

Google is also promising a Pro version of Gemini 3.5, which it describes as its most powerful model, while continuing work on Gemini 4. For now, Gemini 3.6 Flash and Gemini 3.5 Flash-Lite are available starting today for Gemini enterprise users and people using the Gemini app. 3.5 Flash-Lite is also set to roll out to Google Search soon.

Recommended reading

Kimi K3 Nearly Matches Fable at a Fraction of the Cost