I Pay $20 a Month for Claude, But a Free Gemini Feature Keeps Pulling Me Back

I pay $20 a month for Claude, and I keep paying. The model is good at the things I do most, which is long-form writing and code review, and I have built enough muscle memory around the workflow that switching costs me real time. But there is a free Google feature that I open at least three times a week, and every time I open it I wince a little at the subscription charge on my card. This is the post I wish someone had written for me six months ago: a candid comparison of when to use which, what each one is bad at, and why I do not think this is going to stay the same for long.

I am going to talk about Gemini’s “Deep Research” feature (a mode in the Gemini app where the model plans a research task, browses the web, and writes a sourced report) and the long-context window (the ability to feed the model a very large amount of text in a single prompt) on Gemini 2.5 Pro, because those are the two things I keep coming back to. I am not going to talk about the chatbot-in-general comparison, because every other person on the internet has already done that and the answers are obvious.

What I actually use Claude for

Most days, I open Claude to do one of three things.

The first is code review. I have a small monorepo (a single repository that holds several related projects) and I like having a second pair of eyes on a pull request before I merge it. Claude is reliable at this. I paste a diff, ask it to flag anything it would question, and I get back three to five things that are usually worth a second look. False positives are rare. I do this on roughly 60% of my merges.

The second is long-form drafting. When I am writing a 1,500-word article, I will often use Claude to outline the structure, draft the parts where I am stuck, and then rewrite everything in my own voice. The model is good at this in part because Claude is a good writer, and in part because I have a long system prompt (a set of persistent instructions that travel with every conversation) that tells it what my voice sounds like. The output is not publishable, but it is good raw material.

Structured data extraction is the third thing. I have a recurring task where I read a vendor’s pricing PDF and turn it into a spreadsheet. Claude is the only model I have used that can read a 40-page PDF and return a clean table with the right row count. Gemini is competitive on this, and GPT-4o is slightly worse, but Claude wins by a small margin.

What I keep going back to Gemini for

The thing I do not get from Claude is free, and it is not the model itself. It is the integration with Google’s products.

I use Gemini Deep Research when I need a report on a topic I do not know well. Last week I needed to understand the current state of EU AI Act enforcement, and I had a choice. I could pay OpenAI or Anthropic for a deep research tool, or I could click one button in the Gemini app and get a 12-page report with citations in 90 seconds. The report was not perfect, and I had to verify the citations manually, but the time savings were 30 to 40 minutes. The fact that it was free, and that I did not have to leave my Google account, made it the obvious default.

The long-context window is the second thing. Gemini 2.5 Pro ships with a 1 million token context (the maximum amount of text the model can read in a single conversation, where roughly 1 million tokens is about 750,000 words or a small library of books) for free, with limits. Claude’s 200K context is plenty for most of my work, but I have run into the ceiling twice in the last six months: once when I pasted an entire book chapter for editing, and once when I tried to load a 600-page vendor manual for question-answering. In both cases I had to chunk the input and lose the global view. Gemini handled both as a single upload.

Workspace integration is the third reason. I live in Google Docs. The Gemini sidebar in Docs is not great as a writer, but it is fine for the things writers actually use AI for in a doc, which is “summarize this section” and “rewrite this paragraph in a more formal tone.” I use it more than I expected to.

The honest comparison

Neither model is universally better. Here is how I would split the workload, given my own usage and no other context.

• Use Claude for: code review on real codebases, long-form writing with a defined voice, structured extraction from messy documents, anything where you need the model to follow a long system prompt and not drift.
• Use Gemini for: research with citations, anything that needs a 1M+ context window, anything that lives in Google Docs or Gmail, quick summaries of files you already have in Drive.
• Use both for: most days.

The interesting thing is that the things Claude is best at and the things Gemini is best at are barely overlapping. I have not found a single task where I tried both and they were close enough that the choice did not matter. They are both good. They are good at different things.

The costs that are easy to forget

The $20 a month for Claude is the easy number to focus on. The harder-to-count costs are on the Gemini side. When I use Gemini, I am trading my data and my queries for the free tier. Google has not been specific about what they use the inputs for, but the privacy policy allows them to use free-tier prompts for model improvement unless you turn off “Gemini Apps Activity” in the settings. I turn it off. If you are doing anything sensitive, you should too.

Context switching is the other hidden cost. I have a tab open for Claude and a tab open for Gemini on most days. I am good at this. Most people are not, and the temptation to use whichever one is open for the wrong task is real. The right tool for the wrong job is worse than the wrong tool for the right job.

Free tier rate limits are the third thing I did not expect. Gemini’s free tier has daily caps on Deep Research and on the long-context window. I have hit the Deep Research cap exactly once, on a day I needed three reports in a row. The cap is documented and reasonable, but it is also invisible until you hit it.

What I expect to change in the next 12 months

I do not think this comparison is going to stay stable. Three things are likely to shift in the next year.

First, both companies will add the other’s best features. Anthropic will ship a Deep Research equivalent, and Google will ship a better coding model. The integration moat (the advantage a product has by being built into tools you already use, like Docs or Sheets) that Google has with Workspace is real, and Anthropic will have to find a different wedge.

Second, the free tier is going to get more restrictive. Google is subsidizing Gemini to win share. The moment the model is competitive without the subsidy, the free tier will shrink. I would not plan a workflow around today’s free tier.

Third, agentic features (tools that let the model take multi-step actions like browsing, clicking, and writing files on your behalf) are going to matter more than raw model quality. Both Claude and Gemini are shipping agent modes, and the winner is going to be the one whose agents are reliable enough to delegate real work to. I do not know which one that will be.

Trade-offs

Neither of these tools is free in time. Claude’s $20 a month is the smaller cost. The larger one is the cognitive overhead of running two workflows, two system prompts, two muscle-memory sets. In our case, that overhead is about 10 minutes a day of “wait, which tab do I use for this.” Your overhead will be different if you commit to one tool and never switch.

Migration time, if you are coming from one and adding the other, is about a week. The hardest single thing is recreating your system prompt and your saved chats in the new tool. The easiest is the basic prompt-and-response loop, which works the same on both.

If you are starting fresh and you only have time to learn one, learn Claude. If you live in Google Workspace and you only have time to learn one, learn Gemini. If you can run both, run both. They are not substitutes. They are complements, and the sum is meaningfully better than either alone.

What I would tell past me

• Do not try to pick one. Picking one and forcing it to do the other’s job is the most common mistake I see in online comparisons.
• Turn off data training on the free tier. This is in the Gemini settings under “Gemini Apps Activity.” It is one click and it is the right default.
• Use Deep Research for breadth, not depth. The reports are good starting points. They are not authoritative.
• Keep your system prompts in a separate doc. When you switch tools, you will be glad you did.
• Re-evaluate in six months. The model that wins this comparison today will not be the one that wins it in 2027.

Bottom line: I keep paying for Claude, and I keep coming back to Gemini. The $20 is worth it for the code review and the writing. The free tier is worth it for the research and the long context. If I had to pick one, I would pick Claude. I do not have to pick one, and neither do you.