Google keeps pushing the Gemini family forward. While waiting for Gemini 3.5 Pro to launch, the company unveiled two new models at once - Gemini 3.6 Flash and Gemini 3.5 Flash-Lite - and shared its plans for the next generation of artificial intelligence.
The updates touch model performance, code generation quality, document handling, and agentic scenarios. Google continues to refine the Gemini lineup, making the models more efficient both for everyday tasks and professional use.
Gemini 3.6 Flash: faster and more efficient
Gemini 3.6 Flash builds on the previous Flash version, incorporating feedback from developers and enterprise customers gathered after the Google I/O 2026 conference.
The main change is the optimization of internal processes. Google reworked the model's reasoning mechanism, so it now handles complex tasks faster, uses computational resources more efficiently, and relies less on unnecessary intermediate steps.
The developers also updated the model's knowledge base - it now includes information up to March 2026, allowing Gemini to account for more recent events, technologies, and changes across various fields.
Coding was another major area of improvement. According to Google, Gemini 3.6 Flash follows instructions more precisely, makes fewer unnecessary code changes, and is better suited for building complex software projects.
Better knowledge handling
Google also reports that the new model version has become better at analyzing information, handling multi-step tasks, and using external tools.
This is especially important for building AI agents and complex automated workflows, where a model needs not only to answer questions but also to independently carry out a sequence of actions to reach a result.
Gemini 3.5 Flash-Lite: a model for high-volume tasks
Alongside Gemini 3.6 Flash, the company introduced Gemini 3.5 Flash-Lite - a model built for scenarios where speed and handling large volumes of requests matter most.
Google recommends using Flash-Lite for:
- document processing;
- information retrieval;
- building AI agents;
- intelligent assistants;
- business process automation.
Compared to the previous version, the model has gotten better at coding, logical tasks, and tool interaction while maintaining high performance.
Despite being the lightweight version, Flash-Lite received notable improvements and is now better suited for building AI services, automating workflows, and processing large volumes of data.
What we know about Gemini 3.5 Pro
The flagship Gemini 3.5 Pro is still in testing.
Google says the model is undergoing final checks and will become available to a wider audience once all testing is complete. It's expected to become the company's main all-purpose solution for the most demanding tasks.
Google is already training Gemini 4
Beyond the new models, Google officially confirmed that pretraining for Gemini 4 has begun.
Details about the next generation haven't been disclosed yet, but the fact that training has started shows the company is already working on the next stage of its language model development.
Amid intense competition with OpenAI, Anthropic, xAI, and other AI developers, Google keeps focusing on improving model quality and efficiency while expanding its AI offerings.
Summary
The Gemini lineup update shows that Google keeps steadily advancing its AI models. The company is improving performance, boosting code quality, refining how complex tasks are handled, and expanding the models' capabilities for everyday and professional use alike.
Gemini 3.6 Flash received notable improvements in efficiency, coding, and handling complex scenarios. Gemini 3.5 Flash-Lite has become a more capable solution for document processing, information retrieval, automation, and building AI agents, all while keeping high speed.
Meanwhile, Google continues testing Gemini 3.5 Pro, expected to become the company's flagship all-purpose model, and has already begun pretraining Gemini 4. This means the Gemini ecosystem's development isn't slowing down, and new capabilities will keep rolling out regularly.
Competition between Google, OpenAI, Anthropic, xAI, and other developers keeps intensifying. The winners are users, who get faster, more accurate, and more capable tools for text, coding, data analysis, and everyday task automation.
Where to try the new Gemini models
- The Gemini app — gemini.google.com: Google's chat with a free usage limit and a subscription for advanced features.
- Google AI Studio — model access for developers and API testing.
- You can't download the models — the weights are closed and not published; Gemini is only available through cloud services.
- In GPTunneL — both models are already in the catalog: Gemini 3.6 Flash and Gemini 3.5 Flash-Lite sit alongside GPT, Claude, and other models in one interface. No VPN needed, local payment methods supported, no subscription required.
How much the new Gemini models cost
In GPTunneL, both models are billed per token:
- Gemini 3.6 Flash — $0.003 per 1K input tokens and $0.015 per 1K output tokens, with a 1M-token context window;
- Gemini 3.5 Flash-Lite — $0.60 per 1M input tokens and $5.00 per 1M output tokens.
There's no subscription — top up your balance and pay only for the tokens you use. Current numbers for every model are on the pricing page.
New Gemini models FAQ
Which of the two should I pick? Gemini 3.6 Flash — for coding, agentic scenarios, and more complex tasks; Gemini 3.5 Flash-Lite — for high-volume, simpler tasks where speed and minimal cost per request matter most.
Are the new Gemini models free? Google offers a free usage limit in the Gemini app, which is enough for light scenarios. In GPTunneL the models run on pay-per-token pricing: no chat limits and no mandatory subscription.
Do I need a VPN for Gemini? The official Gemini app isn't available in every region and requires foreign payment methods. In GPTunneL both models work right in your browser — no VPN needed, with local payment methods supported.
Try it yourself
The best way to choose between the two is to run the same working scenario through both: open Gemini 3.6 Flash in GPTunneL and compare the result with Flash-Lite on quality, answer length, and price.