Summary
Google has released new AI models designed to cut costs and speed up work for businesses using automated software agents. The new Gemini 3.6 Flash and 3.5 Flash-Lite models focus on reducing the number of tokens—the basic units of text AI processes—needed to complete tasks. This makes running AI agents in production environments cheaper and faster, especially for companies that rely on thousands of automated workflows every hour.
Main Impact
The biggest change is that Gemini 3.6 Flash uses up to 65% fewer output tokens than its predecessor, Gemini 3.5 Flash, in certain tests. For businesses running AI agents around the clock, this directly lowers costs and speeds up response times. Google is also adding a built-in computer-use tool, which lets AI agents interact with operating systems without needing custom software. This simplifies development for engineering teams building background agents for tasks like coding, document processing, and security patching.
Key Details
What Happened
Google announced three new models this week: Gemini 3.6 Flash for coding and reasoning, Gemini 3.5 Flash-Lite for high-volume work, and a restricted Gemini 3.5 Flash Cyber for fixing security flaws. The company also added a client-side computer-use tool directly into its Gemini API and enterprise platforms. This tool lets AI models operate on top of an operating system without extra intermediary software.
Important Numbers and Facts
Gemini 3.6 Flash costs $1.50 per million input tokens and $7.50 per million output tokens. On the DeepSWE benchmark, it scored 49% success, up from 37% for the older model. On MLE Bench, it improved from 49.7% to 63.9%. The model also scored 1421 on Google's GDPval-AA v2 test, compared to 1349 for Gemini 3.5 Flash. Gemini 3.5 Flash-Lite runs at 350 output tokens per second, the fastest in the 3.5 series. It costs $0.30 per million input tokens and $2.50 per million output tokens. On the GDM-MRCR v2 long-context test, it achieved 72.2% success, up from 60.1%.
Background and Context
Enterprise AI agents are software programs that perform tasks automatically, like analyzing documents, writing code, or searching databases. Each task requires the AI to process tokens—pieces of text or data. The more tokens a model uses, the more it costs and the slower it runs. For companies running thousands of agent tasks per hour, even small savings in token usage can lead to big cost reductions. Google's new models aim to balance reasoning ability with efficiency, so businesses can use AI agents without breaking their budgets.
Public or Industry Reaction
Several companies have already started using the new models. Figma integrated Gemini 3.6 Flash into its prototyping tools, with its Director of Product Engineering saying it speeds up design work without losing quality. Legal tech platform Harvey and research tool Hebbia use the model for processing financial filings and creating draft reports. Google also reports improved safety features, with better resistance to misuse in areas like chemical or biological threats, without blocking normal requests.
What This Means Going Forward
For businesses, these models make AI agents more practical for everyday use. The lower token costs mean companies can run more agents for the same price. The built-in computer-use tool reduces the technical work needed to set up agents. However, the restricted Gemini 3.5 Flash Cyber model is only available to governments and vetted partners, which limits its use for general security work. Google is also working on Gemini 3.5 Pro and the next Gemini 4 architecture, so more updates are likely coming.
Final Take
Google's new models show a clear focus on making AI agents cheaper and faster for real-world business tasks. By cutting token usage and adding built-in tools, the company is helping engineering teams build more efficient automated workflows. The trade-off between reasoning power and cost is becoming less of a problem, which could lead to wider adoption of AI agents across industries.
Frequently Asked Questions
What are tokens in AI models?
Tokens are small pieces of text or data that AI models process. For example, the word "hello" might be one token, while a longer word like "understanding" could be split into multiple tokens. The more tokens a model uses, the more it costs and the longer it takes to respond.
How do these new models save money for businesses?
Gemini 3.6 Flash uses up to 65% fewer output tokens than the previous version in some tests. This means it can complete tasks using less data, which directly lowers costs. The cheaper Flash-Lite model also handles simple tasks at a much lower price, so companies can use it for high-volume work without spending too much.
Who can use the Gemini 3.5 Flash Cyber model?
This model is only available to governments and vetted partners through a pilot program. Google restricts access to prevent the model from being used to create harmful code. It is designed to help security teams find and fix vulnerabilities in software, not for offensive purposes.
