Context Windows, Compared for Classroom Use

"Context window" is the most important AI spec a teacher never hears explained. Vendors quote it in tokens — 400,000 here, a million or two there — as if teachers moonlight as machine-learning engineers. They don't. So here's the translation that matters: the context window is how much the model can hold in its head at once. Everything you've uploaded, everything you've typed, everything it has said back — all of it has to fit inside that window, or the oldest parts fall out. For a teacher, that single number quietly decides whether you can hand the model a whole class set of essays or whether you're feeding them in one at a time.

That's not a small distinction. It's the difference between "grade this stack against my rubric and flag the three students who misread the prompt" and "paste one essay, get feedback, clear it, paste the next." Same task, wildly different afternoon.

What a token actually is, in classroom terms

A token is roughly three-quarters of a word — punctuation and word-pieces included. You don't need to do the arithmetic; you need a mental yardstick. A rough, usable rule: 1,000 tokens ≈ 750 words ≈ a page and a half. So a 400,000-token window holds somewhere around 300,000 words — call it several hundred pages. A 1-to-2-million-token window holds millions of words — a whole shelf.

The reason this feels abstract is that most teacher use never bumps the ceiling. Drafting a worksheet, rewriting an email to parents, brainstorming a hook for tomorrow's lesson — these are small. You could do them inside a window a fraction of any of these sizes and never notice the limit. The window only becomes the story when you start feeding the model large, whole things: a novel, a semester of student work, a full curriculum unit, a stack of IEPs, a district policy document you need summarized against a specific question.

Where the numbers actually differ

Among the classroom AI tools, the published context figures spread out considerably. Per the 2026 comparisons at Tech-Insider and Tactiq:

A note on that last row, because accuracy matters here: OpenAI markets ChatGPT for Teachers on its capabilities — unlimited GPT-5.1 Auto, file uploads, data analysis, deep research — rather than on a headline token count. So the honest comparison isn't "OpenAI's number vs. Google's number." It's that Gemini and Copilot have made their context windows a marketed spec, and Gemini's 1M–2M figure is the largest of the group. If your workflow genuinely depends on holding an enormous amount of text at once, that's a real, checkable Gemini advantage — and it's the kind of edge we weigh across all four vendors in The Four-Way AI Vendor Race for the Classroom.

A 400K window holds a few hundred pages. A 1–2M window holds a whole semester of student writing. For most lesson-planning that gap is invisible — but the day you try to grade a class set in one pass, it's the only spec that matters.

The class-set test

Here's the concrete scenario that separates the windows, because it's the one teachers actually hit. You've got 30 argumentative essays, roughly 1,000 words each. That's ~30,000 words of student writing, plus your rubric, plus your instructions — call it 45,000 tokens all in. Every window in the table above swallows that comfortably. So for a single class set, context size is a non-issue, and you can absolutely upload the stack.

Now scale it. You want end-of-semester analysis: five assignments across three classes, ~450 essays, plus rubrics and your running notes on each student. Now you're in the hundreds of thousands of words — and that's where a 400K window starts to strain while a 1–2M window keeps holding it all in one conversation. The teacher with the bigger window asks "which students improved most between assignment one and assignment five?" in a single pass. The teacher with the smaller window batches it, class by class, and stitches the answers together.

Neither is wrong. But it reframes what the spec is for. Context size isn't a bragging number; it's a ceiling on how much you can reason over at once, without breaking the work into chunks and losing the connections between them.

The practical rule for teachers

Three things worth internalizing, and then you can stop thinking about tokens.

First, for everyday work, ignore the number. Worksheets, emails, single essays, lesson drafts — none of them come close to any of these windows. Chasing a bigger context window to write a spelling quiz is buying a freight truck to carry a backpack.

Second, the window resets when you start a new chat. A common beginner frustration — "it forgot what I told it an hour ago" — is often not a context-size problem at all but a new-conversation problem, or the model's separate long-term memory feature, which is a different thing. Keep a single big task in a single conversation.

Third, when the task is genuinely large and whole — a semester of writing, a full curriculum, a long document you're interrogating — the context window becomes the spec that decides whether you upload it all or feed it in pieces. That's the moment to care, and the moment Gemini's headline figure earns its keep. For most teachers, most weeks, that moment is rare. But when it comes, no amount of clever prompting substitutes for a window big enough to hold the whole thing. And the size of that ceiling is one more reason the classroom AI market hasn't collapsed into a single winner — the vendors are optimizing for different jobs, which is exactly the argument in The Four-Way AI Vendor Race.

Part 77 of 100 in the ChatGPT for Teachers series. Previously: The Real Price of "Free": A Cost Comparison. Next: Why OpenAI Chose Free Over a Freemium Tier. Browse more builder insights or explore AI skills for education at aiskill.market.