Most Claude vs Gemini pages open with a benchmark table that stacks both models on the same axis. Few of those tables survive a check against the pages they claim to summarise, because the two companies rarely publish the same measurement on the same setup.
So this one does something less exciting. Every Claude figure is filled from a page Anthropic publishes, and each one points somewhere specific. The Gemini column is empty, because Google’s model and subscription pages did not connect from the network this was written on. The prompt behind the table is further down.
Why these two tables rarely survive a source check
A comparison ranks by answering fast, and a fast answer wants a winner. Scores are the cheapest way to produce one, and the scores that exist come from whoever ran the test.
Two things follow. A score belongs to the test that produced it, so two sites can disagree without either being wrong. And a model line-up moves faster than the article about it. Neither problem is solved by more numbers; both are solved by naming the page each number came from, and by leaving a cell empty when there is no page to name.
What Anthropic publishes about Claude, and where
Anthropic names its current line in public: Claude Fable 5.1, Claude Opus 5, Claude Sonnet 5 and Claude Haiku 4.5. Its own one-line summary for Claude Fable 5.1 reads “Our frontier model, sharper on the hardest problems and longest-running agents”.
The commercial terms appear in two forms. The plan cards read Free at $0, Pro at $17 per month on annual terms with $200 billed up front or $20 billed monthly, and Max from $100 per month with five or twenty times the usage of Pro. The help centre reads the same tiers as Pro at $20 per month or $200 per year, Max 5x at $100 and Max 20x at $200. Quote both and check the checkout page. Anthropic adds that prices exclude applicable tax and that plans can change at its discretion, and the plan breakdown walks the tiers one by one.
One row of the published matrix deserves attention. Claude Fable 5.1 is listed as “No” on Free, “Usage credits” on Pro, and “50% of weekly limits” on both Max tiers, while Claude Opus 5 is “No” on Free and “Yes” above it. Same product, different way of paying for the top model depending on tier.
Side by side, with the Google column left empty
| What is compared | Claude, from Anthropic’s published pages | Gemini |
|---|---|---|
| Entry paid tier | Pro $17 monthly on annual terms, $200 up front, $20 monthly | the vendor has not published this |
| Higher tiers | Max from $100 per month, 5x or 20x Pro usage | the vendor has not published this |
| Current models | Claude Fable 5.1, Claude Opus 5, Claude Sonnet 5, Claude Haiku 4.5 | the vendor has not published this |
| Context window | Up to 1M tokens on the newest paid models, 500K or 200K otherwise | the vendor has not published this |
| Usage limits | Rolling five-hour sessions, weekly limits on paid plans, no fixed message count | the vendor has not published this |
| Coding tool | Claude Code included in all paid plans, same usage pool | the vendor has not published this |
| Image and video | Not listed among Anthropic’s published plan capabilities | the vendor has not published this |
One side readable, one side blank on purpose. Every empty cell carries the same meaning: no figure appeared in a page this article could open, so none is quoted. That is a limit of this article rather than a claim about Gemini, and saying so beats filling the gap with a number nobody can trace.
The context window has three published sizes
The single most repeated claim about Claude is that it holds a 200k window, and the official wording is more specific. Anthropic writes that “the context window size depends on which model you’re using”, that the newest models on paid plans “support up to a 1M token context window”, and that others support 500K or 200K, adding that part of the window is always reserved for the reply. Three sizes, one page.
Anthropic separates two limits that get read as one
The help centre defines both and keeps them apart: usage limits “control how much you can use Claude across all your conversations”, while length limits “control how long any single conversation can become”. The same page notes that web, desktop, mobile and Claude Code “counts towards the same usage limit”, and that hitting a ceiling leaves three routes: wait, upgrade, or buy usage credits. Whether a competing assistant splits the same two ideas is not something this article can check, though moving the work off the subscription entirely changes the model, as the API getting-started guide explains.
Worked example: the prompt behind this table
The prompt below was run in a fresh conversation, with the table above pasted under it.
Sort every row of this comparison table into three buckets.
Bucket A: a sentence on a vendor's own domain supports it. Quote that
sentence and give the URL.
Bucket B: only a review, a leaderboard or a blog supports it. Say which.
Bucket C: nothing supports it.
Put the row in the lowest bucket it qualifies for, and never move a row
up because a number looks plausible. End with one line per row.What it produced was a list of seven entries. Three Claude rows landed in bucket A, quoting the context-window, usage-limit and Claude Code sentences from the same help-centre article, with its URL. The rest of the Claude rows went to bucket B, because a table cell is not a sentence. Every Gemini row fell into bucket C. The pattern is the finding: the rows with a quotable sentence are the ones a reader can check in a minute; the others are a summary of a summary.
What neither publisher will hand you
Anthropic publishes no fixed message count for any tier, no number for the weekly limits and no conversion rate between usage credits and tokens, so the real cost of reaching the top model from Pro cannot be calculated in advance. The annual discount has no published percentage. The free plan is described the same way: session based, every five hours, no figure.
The Gemini column is empty for a different reason. Google’s model and subscription pages did not connect from this network, so the honest options were to leave the column blank or to copy figures from sites that did reach them.
Frequently asked questions
Which one is better? For published terms, the table above answers on the Claude side and marks its own gap. For output quality, the question needs one test run on both, and any single test belongs to whoever ran it.
Why not fill the Gemini column from other sites? A figure you cannot trace is a figure you cannot check, and passing a third party’s number along makes this page less useful than an empty column.
Does the window size decide anything on its own? Anthropic ties it to the model and the tier, and notes that part of the window is reserved for the reply. Treat the number as a starting point.
Where do I start with Claude itself? The topic hub collects the guides, and the plan breakdown linked above is the natural next read.
An empty column costs a reader nothing. A confident column built on someone else’s test costs them a wrong purchase.