Google has rolled out two new additions to its Gemini lineup: Gemini 3.6 Flash and Gemini 3.5 Flash-Lite. These models are basically built for developers and big companies who need faster AI, sharper reasoning, and lower costs when dealing with heavy workloads.
That's not all, though. Google also said Gemini 4 is already in pre-training. They are calling it their most ambitious AI project yet. There is no release date, but it’s clear the next big Gemini generation is already in the works.
So, what makes Gemini 3.6 Flash interesting?
Well, it takes what Gemini 3.5 Flash did and pushes things further—coding gets a boost, it understands multimodal inputs better, and it’s smarter with knowledge-based tasks. Google says the new model spits out answers more efficiently, using 17 per cent fewer output tokens in the Artificial Analysis Index tests. On the DeepSWE benchmark (that’s from DataCurve), output tokens drop by as much as 65 per cent. That directly reduces how much it costs to run.
And that efficiency does not stop there. Google claims Gemini 3.6 Flash needs fewer reasoning steps and fewer tool calls to handle multi-step, complex tasks. That’s great if you are running AI agents or want to use it for business apps—less waiting, less processing cost.
If you are wondering about price: it’s USD 1.50 (about Rs. 144) for one million input tokens and USD 7.50 (about Rs. 720) for one million output tokens.
How does it stack up on benchmarks?
Google shared some numbers:
- DeepSWE: 49 per cent (up from 37 per cent for Gemini 3.5 Flash)
- MLE Bench: 63.9 per cent (up from 49.7 per cent)
- OSWorld-Verified: 83 per cent (up from 78.4 per cent)
It’s not just about benchmarks, either. The model handles various multimodal tasks, like
- Parsing documents
- Analyzing charts and data
- Generating reports
- Computer-based functions through the Gemini API and Enterprise suite
Gemini 3.6 Flash has stronger safety protections
Plus, Gemini 3.6 Flash has stronger safety protections, especially around cyber threats and anything involving chemical, biological, radiological, or nuclear domains (that’s the CBRN stuff).
Gemini 3.5 Flash-Lite:
Now, for Gemini 3.5 Flash-Lite. This one’s made for people who need speed but want to keep costs low. Google calls it the fastest and most affordable option in the Gemini 3.5 lineup. It can shoot out up to 350 output tokens per second, making it a good fit for the following:
- AI search
- Processing documents
- Handling high-volume chatbots
- Automating agent workflows
- Latency-sensitive business apps
Gemini 3.5 Flash-Lite pricing starts lower: USD 0.30 (about Rs. 29) per 1 million input tokens and USD 2.50 (about Rs. 240) per million output tokens.
Developers can pick different reasoning strengths depending on how tricky their workloads are, trading off between price, speed, and intelligence.
There is more, as Google also announced Gemini 3.5 Flash Cyber, a security-focused model. This one’s trained to spot software vulnerabilities, check security flaws, and help patch code. It works with Google’s CodeMender security agent. But you probably will not get your hands on it, as it is not for public use yet. Only governments and certain trusted partners get access, and even then, it's just a pilot for now.
If you want to try out the new models, they are available now through the Gemini API, Google AI Studio, Android Studio, and Gemini Enterprise Agent Platform. Gemini 3.6 Flash also works with Google Antigravity and the Gemini Enterprise app, and both are rolling out to the Gemini app soon. Gemini 3.5 Flash-Lite is getting plugged into Google Search as well.
Lastly, Gemini 3.5 Pro is not quite ready for everyone, as it is still being tested with select partners, but a wider release is planned once it’s good to go.
In the bigger picture, Google’s already hard at work on Gemini 4. They have started pre-training, setting the stage for their largest and most ambitious AI yet. No firm launch date, but the next step for Gemini is definitely underway.
