Google's paid AI agent just went free, and India got it on day one
I build automation agents as a hobby, so I pay attention when a big company takes a paid agent feature and hands it to free users. On November 17, Google did exactly that. Gemini Tasks, the agent mode previously locked b
I build automation agents as a hobby, so I pay attention when a big company takes a paid agent feature and hands it to free users. On November 17, Google did exactly that. Gemini Tasks, the agent mode previously locked behind the AI Pro and AI Ultra subscriptions, started rolling out to free Gemini accounts in six countries: the US, Japan, South Korea, Canada, Australia, and India.
If you open Gemini on your phone or the web, there is a toggle above the prompt box that says Chat. Flip it to Task. Instead of walking you through an answer step by step, Gemini takes the whole job and works on it in the background. You can hand it a trip plan, a research topic for a school project, or a recurring morning briefing. It runs on Gemini 4 Hydrogen, the fast model Google also made the default in Search AI Mode on November 12.
The part that surprised me: it can now call local businesses using Google Duplex, pay through Google Pay, receive texts over RCS, and accept forwarded emails in Gmail. Those features are US-first for now. It can also sign into websites through Google Password Manager without exposing your passwords outside Google's ecosystem. I have mixed feelings about that last one. Convenient, yes. But handing an agent your logins is a lot of trust, and I would want to watch what it does before letting it loose on anything important.
For students in India, the free part matters more than the agent part. This was a paid feature two weeks ago. Now a class 10 kid with a free account can hand off research tasks while doing homework. That is the actual story here.
Source: Android Central's report on the rollout, based on Josh Woodward's November 17 announcement. Try it: gemini.google.com
The launches nobody is talking about
While everyone argued about Gemini, a few quieter releases landed this month. These are the ones I would actually try.
MiniMax H3-Image: open weights, native 4K
On November 10, MiniMax released H3-Image, a 33-billion-parameter image model that generates native 4096x4096 images in about 10 seconds. No upscaler pass. One checkpoint handles text-to-image and editing with up to 16 reference images. The weights are public on Hugging Face, and the API costs roughly $0.022 per 1K image and $0.069 per 4K image.
The catch worth knowing: the community license excludes the US, EU, UK, and South Korea. India is not on that list, so for developers here this is one of the strongest free-to-download image models available right now. Read the license before building anything commercial.
Official announcement: minimax.io/blog/minimax-h3-image. Weights: huggingface.co/MiniMaxAI/MiniMax-H3-Image
UnoRouter: one free API key for every major model
If you have ever juggled separate API keys for GPT, Claude, Gemini, and DeepSeek, UnoRouter is worth a look. It is an OpenAI-compatible gateway: one key reaches all of them, with free models to start. It showed up quietly in developer directories this week. For anyone building bots or agents on a budget, that is a real shortcut. I have not tested it in production yet, so treat it as an experiment, not infrastructure.
Profile and details: ToolsAIApp profile
Google Vids: free AI video clips
Google added free AI video generation to Vids on November 5: three clips a month, up to 15 seconds each. Small allowance, but it is free, and it sits inside a tool that already turns slides into video. For a student making a project presentation, three clips is enough. Try it: vids.google.com
What I am doing with this
I am testing Gemini Tasks against my own automation scripts this week. If a free agent can handle the boring research steps, my scripts get shorter. I will write up what breaks.
Which one are you trying first, the free agent or the 4K image model?
Drafted with AI assistance (#ABotWroteThis). Every fact and date was checked against the linked sources.
Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.