- Published on
A Cheaper, More Token-Efficient Flash Tier Built for Agentic Workloads
- Authors
- Name
- DeepHorus
Google has released three new Gemini models designed to optimize cost and speed for production agentic workloads. The lineup includes Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber, all part of the Flash tier that prioritizes token efficiency and low latency over maximum reasoning depth. These models aim to meet the needs of developers building high-volume agents.
The Flash tier is tuned for speed and cost-effectiveness, making it ideal for agentic tasks requiring reliable performance at scale. Building on previous iterations, the new models offer significant improvements in efficiency, quality, and specialization. Google positions them as workhorses for coding, knowledge work, and multimodal tasks.
Gemini 3.6 Flash uses 17% fewer output tokens than its predecessor on the Artificial Analysis Index, with up to 65% reduction on the DeepSWE benchmark. It is priced at $1.50 per 1 million input tokens and $7.50 per 1 million output tokens, down from $9.00. Quality scores improved across benchmarks: DeepSWE 49% versus 37%, MLE Bench 63.9% versus 49.7%, and OSWorld-Verified 83.0% versus 78.4%.
Gemini 3.5 Flash-Lite achieves 350 output tokens per second at $0.30 per 1 million input tokens and $2.50 per 1 million output tokens. It outperforms prior models on Terminal-Bench 2.1 and GDM-MRCR v2, and even beats the older Gemini 3 Flash on some evaluations. Developers can configure thinking levels from minimal to higher, balancing cost and complexity.
Gemini 3.5 Flash Cyber is specialized for finding and patching software vulnerabilities. It powers CodeMender, Google's code-security agent, running multiple parallel agents to explore execution spaces. On Big Sleep evaluation, it found 55 unique confirmed issues in the V8 JavaScript engine, surpassing mainline 3.5 Flash and 3.6 Flash.
Early customers such as Hebbia and Harvey have reported gains in document parsing, chart analysis, and report drafting using Gemini 3.6 Flash. Google has also enhanced Frontier Safety safeguards for 3.6 Flash to cover CBRN and cyber-offense misuse. These releases significantly reduce the cost per agentic task while improving performance.
Looking ahead, these models could accelerate adoption of AI agents in production environments. With built-in computer use and specialized capabilities, Google aims to solidify its position in the agentic AI space. Developers now have more efficient and tailored tools for building scalable, cost-effective agents.
