Google updates Gemini Flash line to slash enterprise agent costs

For teams building agents that perform thousands of operations hourly, parameter count is secondary to efficiency. Gemini 3.6 Flash focuses on multimodal reasoning and coding, reporting a 17 percent reduction in output tokens compared to its predecessor. According to data from the Artificial Analysis Index, the model achieved a 63.9 percent success rate on the MLE Bench, up from 49.7 percent. Pricing is set at $1.50 per million input tokens and $7.50 per million output tokens, a structure intended for continuous reasoning loops.

Early adopters are already integrating the model into specialized workflows. Figma uses 3.6 Flash to accelerate design iterations, while Harvey and Hebbia leverage its multimodal capabilities to parse financial filings and embedded charts. Google has also streamlined the development process by integrating a client-side computer-use tool directly into the Gemini API, eliminating the need for custom middleware when agents interact with operating systems.

For high-volume, low-complexity tasks, Google introduced Gemini 3.5 Flash-Lite. With a throughput of 350 output tokens per second and pricing starting at $0.30 per million input tokens, it serves as a lightweight alternative for document processing. Meanwhile, the company is piloting Gemini 3.5 Flash Cyber, a restricted model tailored for vulnerability remediation. Distributed solely to governments and vetted partners, this variant utilizes a cross-checking mechanism within the CodeMender security agent to validate code patches before human review.