Claude vs Gemini: price, context and limits compared

Claude, from Anthropic, and Gemini, from Google, are both offered as general-purpose assistants and API models, and both vendors publish the figures this page compares: price per unit of text, how much a model can take in at once, and how much it can produce in a single reply. What this page does not do is say which model is better. It sets the comparable facts side by side and leaves the judgement to you.

What you can actually compare

Pricing for both models is metered — a rate for the text you send in, a separate and higher rate for the text generated back. That input-versus-output split, and the fact that output costs more per unit than input, holds across this category generally rather than being a quirk of either vendor. One detail worth flagging for Gemini specifically: Google’s pricing for this model is tiered by prompt length, so the rate itself can shift depending on how much text you send — the table above reflects that structure rather than a single flat number.

Context window — how much text a model can hold in view for one exchange — is the second point of comparison, and the two vendors describe theirs differently in their own documentation, which the table resolves into a common format. Output limit, the cap on how much a model can write in a single response, is the third axis, and again it’s a separate figure from context, not derived from it.

When the difference matters

At low, everyday volumes, none of this changes much — a few queries a day are cheap regardless of vendor. The gap opens up under volume: a workflow generating a high number of long responses will feel a pricing structure much more than one that fires off short, occasional questions, and Gemini’s tiered pricing in particular means the cost profile can change as your prompts grow.

A large context window earns its keep when the task genuinely needs everything held in view at once — a long document, an extended conversation history, a big batch of reference material — rather than being fed in stages. Short, self-contained tasks rarely test this limit either way.

Output limits matter most when a single response needs to be long by itself: a full draft, an extensive breakdown, a large chunk of generated code, as opposed to a short answer or summary.

What this page won’t tell you

This page is about published price and capacity, not about which model reasons better, writes more naturally, or handles a given task with more nuance. Those comparisons are actively contested, change with every release, and depend on exactly what you’re asking the model to do — none of which a static comparison page can responsibly settle. Independent public evaluation leaderboards are the better place to look if capability, rather than cost or capacity, is what you’re weighing.

It’s also worth saying plainly: Around is built using Claude. That’s exactly why we don’t rank models here — we’d be judging our own supplier, so instead we stick to figures each vendor publishes and let you draw your own conclusions.

For the current figures side by side, including other models beyond these two, use the interactive comparison tool.