Gemini review: The free tier is best-in-class, if you can figure out which Gemini you're using

The first time we noticed Gemini was when we needed to find a specific clause in a 280-page master service agreement someone swore was "definitely in there somewhere." We pasted the whole PDF into Gemini 2.5 Pro on a Saturday morning. "Find any clause that limits our right to use the work product for training ML models." Twenty seconds later: the clause, page 187, with surrounding context and a note that a related limitation also appeared on page 213. Command-F would have taken an hour.

That was three months ago. Gemini sits in our regular rotation with Claude and ChatGPT now. The question worth answering is whether it deserves more than a rotation slot, and the harder question: which Gemini we're even talking about, since Google ships three overlapping versions of the same product under names that change every few months.

The free tier is genuinely class-leading

Most AI assistants give you a stingy free tier built to push you toward Pro. Gemini's is the rare exception. You get 2.5 Flash with a million-token context window, Imagen 3 image generation without a credit limit you'll hit in normal use, web-grounded research with citations, and NotebookLM, which is itself one of the best AI products Google ships. None of it asks for a payment method.

If you're trying to figure out whether AI assistants are useful for you, Gemini is the right starting point, even if you end up paying for Claude or ChatGPT. ChatGPT's free tier feels like a demo. Gemini's feels like a product.

Long context that actually works

The million-token context isn't a marketing number. Drop a full repository into Gemini 2.5 Pro and ask about a function defined in one file and called from another - it gets it. Paste a year of meeting transcripts and ask for the through-line - it finds it. Needle-in-haystack performance is closer to Claude's than ChatGPT's, and on certain document-comparison tasks Gemini is the only one of the three that doesn't lose track of which document said what.

Long contract review: paste the whole thing in, no chunking. Comparing two RFP responses: drop both, ask for a side-by-side. Onboarding to a new codebase: paste the source tree, ask questions. You stop budgeting context prep into the task.

NotebookLM

Buried inside the Gemini ecosystem is NotebookLM, a research synthesis tool where you upload documents, audio, and YouTube videos, and the model grounds itself only in those sources. Ask a question, get an answer with citations back to the specific paragraphs that support it. No outside information. No facts hallucinated from training data.

We use it for client research, learning new technical areas, and combing through depositions. It also generates podcast-style audio summaries of your notebook on demand, which sounded gimmicky to us until we listened to one and found ourselves learning faster than reading.

For reasons we don't understand, NotebookLM gets approximately no Google marketing.

The product sprawl problem

When you say "Gemini" you might mean the gemini.google.com chat app, Google AI Studio, NotebookLM, Imagen, Veo, the API, the Workspace integration, or any of the model variants - 2.5 Pro, 2.5 Flash, Flash Lite, Deep Think, Live, Advanced. New versions ship every few months and old names linger in URLs and tutorials.

The default model in the consumer app isn't the smart one. Ask a hard question without touching the model picker and you get a 2.5 Flash answer that's noticeably worse than what 2.5 Pro would give. New users don't know to switch. We've watched friends conclude "Gemini isn't very smart" because they never left the default.

ChatGPT has this problem too in theory. In practice it routes hard questions to better models automatically. Claude has one default that's good. Gemini makes you do the work of being your own model router.

Where it loses to ChatGPT and Claude

Gemini is more cautious and more hedge-prone than either. Ask for something in a specific tone, you get a watered-down version that won't take a position. Ask for a summary of a controversial event, you get five paragraphs about complexity. Both of those are sometimes the right move; Gemini gives you them by default.

The mobile apps lag the web product noticeably. Features land on web first and may take months to reach iOS or Android. ChatGPT's mobile experience is years ahead in both polish and breadth.

Workspace integration, with a catch

Gemini reads your Gmail, Drive, and calendar if you're signed into a Google account. For anyone already in Workspace this is genuinely useful: ask it to draft a reply to a client email with context from a Drive document and the lookups happen automatically.

The catch is the deepest Workspace integration is gated behind a separate Workspace subscription that doesn't bundle with Gemini Pro. You can end up paying twice. It's also off by default in consumer Gmail for privacy reasons, with a toggle most people never find.

Image and video, bundled

Imagen 3 produces some of the best AI-generated images we've tested, especially photorealistic portraits and complex compositions. It's free inside Gemini and beats paying for a separate Midjourney or DALL-E subscription for casual use. Veo, Google's video model, ships in the same product for paying users - enough for the occasional 8-second clip in a slide deck without paying Runway or Pika separately.

Should you use it?

Yes, especially if you don't pay for AI today. The free tier is best-in-class, NotebookLM alone is worth the signup, and the long-context performance is the best we've used on the work that needs it.

If you already pay for Claude or ChatGPT, Gemini complements rather than replaces them. Keep it open for document-heavy work and for NotebookLM.

We end up reaching for Gemini more than we'd predicted. The product would be unambiguously top-tier if Google would pick one default model and stop launching adjacent surfaces with overlapping names every quarter. They probably won't. Use it anyway.