I downgraded to free Gemini and ChatGPT for a week, and only one subscription is worth keeping

Published Nov 2, 2026, 2:00 PM EST Nolen began their writing career in 2019, with three years dedicated to editing the Creative section at MakeUseOf. Their expertise lies at the crossroads of technology and creativity, covering areas like photography, video editing, and graphic design. Outside of work, you'll often find Nolen diving into a good book, writing their own stories, or playing video games. Free AI plans got quite complicated in October. OpenAI and Google both reworked what non-paying accounts get within days of each other, with OpenAI moving free users onto its newer GPT-6 model while Google cut free Gemini back to its smallest one. Since they're sitting on such different parts of the free AI spectrum now, it makes it a little harder to answer the question of whether a subscription is worth it, or even to compare their free tiers. Dropping to free Claude a few months ago showed me that the things I'd actually miss were Opus and Claude Design rather than only the extra usage. So I wanted to see if there would be extra surprises when dropping down to the free tiers of the other subscriptions I'm currently paying. It was quite a mixed bag with some unexpected strengths. What each free plan actually gets you The fine print matters more than the pricing page Both free tiers look promising when you look at their pricing pages, but the details that actually matter for real use are a little further down. ChatGPT started rolling GPT-6 out to free accounts in October, specifically GPT-6 Luna, the fastest and lowest-cost model in that family. But there's no model picker, ChatGPT chooses your model for you, and you get a Think button that gives the models more time on harder questions. Gemini went the opposite way. Since October 9, free accounts have lost the regular Flash and Pro models and only get Flash-Lite which is Google's smallest model. Auto mode, the default, sends most prompts to Flash-Lite and can route harder ones up to Flash or Pro, but it never tells you which one actually responded. The limits work differently too. ChatGPT's text chat is unlimited now, while uploads, image generation, and voice each have their own caps, and uploading is pretty tight at only three files per day. Gemini doesn't count messages at all. It uses an allowance that refreshes every five hours until you hit a weekly limit, so generating images eats into it faster than chatting does. Context is where it really starts to set itself apart from the subscriptions, and it stings a little. Free ChatGPT holds about 27K tokens in a conversation, roughly 12 pages of text, and free Gemini gets 32K. Being used to Claude's 200K, this truly feels like peanuts. But then again, it's about the same as my local LLMs - the only difference is I don't use my local LLMs as much as cloud AI, and I never expected them to handle the complex multi-step stuff. At least Gemini still includes Canvas and Deep Research on the free plan, and ChatGPT keeps a limited version of deep research too. The agentic tools are largely paywalled though, since Gemini's Agent Mode needs a paid Google AI plan while ChatGPT only offers limited Codex and Work access through its desktop app. But this was mainly about their chatbots and not their agents. What a week of free ChatGPT and Gemini looks like Same tasks, very different results I didn't do anything special with ChatGPT or Gemini and just continued to use them as I normally would, so all the stuff I'd throw at my paid plans, with a handful of prompts kept identical across both so neither got an easier ride. Out of the gate, ChatGPT was the one I trusted faster because anything involving current information came back with tappable citations sitting next to the claims. It was also careful to separate what it actually knew from what it was estimating, and more than once it asked me a follow-up question instead of guessing. Gemini lets free users pick a thinking level, and the default, Low, was pretty rough. So I did actually bump it up to Medium and High for the rest of the week, which of course dug a little more into my context budget. It did help a little, but the answers still stayed short and a bit underwhelming. Unless it decided to search, it leaned on older knowledge. Even when it did search, its reasoning sometimes contradicted numbers it had just given me. When it came to building stuff, my premature opinion changed. Gemini's Canvas is genuinely good on free, and it tends to go beyond the brief with extra interactivity, plus there's a code view and export button. It's looser with instructions though, so a request for an image could come back as a small vector graphic instead, for example. ChatGPT stuck to what I asked more literally, producing working interactive previews and actual image files, although it also generated an extra image nobody asked for, which is just wasteful on a plan with such a heavily capped allowance. Design-wise, both default to the same dark, glassy blue-and-purple look that seems to be every AI's idea of modern. I'd have liked one of them to try something else, though I'm probably the only one bothered by that. The ceiling always shows up eventually The limits you might only notice after a few days Free ChatGPT's less obvious limit is memory. Unlimited messages sounds like the better deal of the two, except its 27K window is smaller than free Gemini's, so long conversations start losing track of their beginning pretty quickly. I remember this being an issue even way back when - after just a couple of prompts on my first free account, ChatGPT would become so incoherent. Files also end up being a problem since the three-a-day cap disappears after five minutes worth of screenshots. The upgrade prompts that pop up out of nowhere are also quite annoying. Gemini's problem is harder to plan around. Because Auto picks the model without saying which one answered, the same question can come back noticeably dated depending on where it got routed. That made it tough to rely on for anything factual, and at least once it recommended an older version of Gemma, Google's own open model, even though Google had released a newer one months earlier - this is a pretty strange vibe for Google. Free got better on one side and worse on the other The weird part is that the chatbot I enjoyed the most on its free version is actually the one I'd cancel. It's precisely because ChatGPT is the more capable free tool that I think I'd be able to get by on its free version. Gemini's free version has been downgraded quite a bit, so to get proper use out of it you kind of have no choice but to get the paid plan. ChatGPT Google Gemini