LLM context window comparison
Paste a prompt or document and see how many tokens it is, instantly — then check it against every major model's context window at once. Fill bars and a fits / too-big verdict show exactly how much room you have, with the model list loaded fresh from OpenRouter.
≈ 0 tokens
≈ 0 words · ≈ 0 pages
| Model | Your text | % used | Fit | |
|---|---|---|---|---|
Token counts use the GPT tokenizer family — other models' tokenizers differ by roughly ±10–20%. Context windows are what providers list on OpenRouter today, fetched live in your browser.
What a context window actually is
A context window is the maximum number of tokens a model can consider in one request — your prompt, documents, conversation history, and the response all share it. Models read tokens, not words: a token is roughly ¾ of an English word, so 1,000 tokens ≈ 750 words ≈ a page and a half. That's why "how many tokens is this document?" is the first question to answer before picking a model — a 128K-token window holds about 96,000 words, or roughly 190 pages, while an 8K window fits closer to 12.
How big are today's context windows?
Windows have grown fast. A few years ago 4K–8K tokens was standard; 32K and 128K became common; and by 2026 the frontier closed models — the current Gemini, GPT, and Claude families — have converged on 1 million tokens, with a handful of specialized models advertising 2M–10M. As a rule of thumb: 8K ≈ a long email, 32K ≈ a short report, 128K ≈ a short book (~190 pages), 200K ≈ a full novel, and 1M ≈ a small bookshelf (~750,000 words). The table above shows the live numbers rather than a snapshot, because providers raise these limits often.
Bigger isn't always better
Fitting isn't the whole story. Models degrade on very long inputs — recall of details buried in the middle of a stuffed window is measurably worse than near the start or end (the "lost in the middle" effect) — and every input token is billed, so a maxed-out window makes each request dramatically more expensive. Aim for comfortable headroom rather than a bare fit, put the passages that matter near the beginning or end, and price a full request with the LLM cost calculator before you rely on a large window.
Fit more in with retrieval, not stuffing
When your content is bigger than the window — or big enough that stuffing it is slow and pricey — the answer is retrieval, not a larger model. RAG pulls only the most relevant chunks per query, so a 500-page manual becomes a handful of passages the model actually needs. Long conversations get summarized or windowed rather than replayed in full. See how chunk size and overlap change what gets retrieved in the RAG chunking visualizer.
Frequently asked questions
- What is a context window?
- A context window (or context length) is the maximum number of tokens a model can consider in a single request. Your system prompt, the user's message, any documents or retrieved context, the conversation history, and the model's response all share that one budget. Exceed it and the request is rejected or the oldest content is dropped.
- How many tokens is my document, and how many words is 128K tokens?
- A token is roughly ¾ of an English word, or about 4 characters, so 1,000 tokens ≈ 750 words ≈ 1.5 pages. A 128K-token window holds about 96,000 words (~190 pages); a 1M-token window holds roughly 750,000 words. Paste your text above for an exact count rather than an estimate.
- Which LLM has the largest context window?
- Frontier closed models have converged on 1 million tokens (Gemini, GPT, and Claude families all offer 1M-class windows in 2026), with some research and specialized models advertising 2M–10M. But the largest window is rarely the right choice: bigger windows cost more per full request and recall degrades on very long inputs. The live table above shows current windows pulled from OpenRouter so the numbers stay honest.
- Does using the full context window cost more?
- Yes. Every input token is billed, so a request that fills a large window is dramatically more expensive than one that sends only what's needed. Filling a 1M-token window on a frontier model can cost several dollars for a single call. Fitting is necessary but not free — check the price with the LLM cost calculator before you rely on a stuffed window.
- Why do models get worse with very long inputs?
- Models exhibit a 'lost in the middle' effect: recall of details buried in the center of a long input is measurably worse than content near the beginning or end. So even when your text technically fits, burying the important passage in the middle of a maxed-out window hurts answer quality. Comfortable headroom — and putting key context at the edges — beats a bare fit.
- How do I fit more into the context window?
- Don't stuff — retrieve. Instead of pasting an entire corpus, use RAG to pull only the most relevant chunks per query; summarize or window long conversation histories; and chunk large documents so you embed and retrieve pieces rather than the whole. Our RAG chunking visualizer shows how chunk size and overlap change what gets retrieved.
- Is this context window tool free?
- Yes — it tokenizes in your browser, needs no signup, and the model list loads live so the context windows stay current. Nothing you paste is sent to a server.
More free tools
- LLM price comparison — Live table of every model's API prices — sort, filter, and cost out your own request size.
- LLM cost calculator — Estimate per-request, daily, and monthly spend from token counts and request volume.
- RAG cost calculator — Full pipeline costs — embeddings, vector storage, and generation — from corpus size and query volume.
- RAG chunking visualizer — Paste a document and watch chunk size, overlap, and strategy change the chunks before you embed.
- Tool call & MCP schema builder — Build function-calling parameters once, copy valid JSON Schema for OpenAI, Anthropic, and MCP.
- AI chatbot ROI calculator — Turn ticket volume, handle time, and deflection rate into monthly support savings and ROI.