Google’s Gemini 3.6 Flash cuts tokens, trails rivals

Google has released Gemini 3.6 Flash, positioning it as a faster, more token-efficient model for everyday workloads. According to the company, it can reduce output-token usage by up to 65% compared with Gemini 3.5 Flash and uses 17% fewer tokens on average.

That efficiency matters as AI-service costs become a routine part of corporate budgets. But early benchmark comparisons give Google less to celebrate: Gemini 3.6 Flash reportedly trails Claude Sonnet 5 from Anthropic and GPT-5.6 from OpenAI on popular tests. It also falls behind Grok 4.5 from xAI in agentic programming tasks, while Google’s pricing is roughly unchanged.

Gemini 3.6 Flash lineup and availability

Google describes Gemini 3.6 Flash as the practical version of its flagship family, prioritizing an acceptable quality-to-cost ratio over benchmark records. The model is available to users of the Gemini app and to enterprise customers.

The release also includes:

Recommended reading

GPT-5.6 Sol leads colored-pencil model showdown

• Gemini 3.6 Flash: up to 65% fewer output tokens than Gemini 3.5 Flash.
• Gemini 3.6 Flash: 17% lower average token usage than the previous version.
• Gemini 3.5 Flash-Lite: the fastest and most economical model in this branch.
• Gemini 3.5 Flash Cyber: a model focused on cybersecurity.

Flash Cyber is available only to governments and trusted partners through CodeMender, an AI agent being piloted to find and fix vulnerabilities. That is a more cautious approach than Anthropic’s rollout of Mythos, which the company described as dangerous enough to restrict to a small group of partners.

Google also says a Pro version of 3.5 is in development and will be the most capable model in this generation. At the same time, the company is training Gemini 4, leaving 3.6 Flash as another step in an ongoing attempt to close the quality gap with Anthropic, OpenAI, and xAI without increasing costs.