AI Context Window Comparison

A model's context windowis the maximum amount of input it can read in a single request, measured in tokens. This tool lists the published window for each model and converts that raw token count into units you can reason about. Every window figure is taken verbatim from the provider's documentation — each row links to its source — and the whole table carries a 2026-07-03 verification date, because these limits change often.

The three real-world conversions use documented averages, kept identical to the site's AI Context Window Calculator so the numbers reconcile between tools:

1. Words = round(tokens × 0.75). OpenAI's tokeniser guidance is that 1 token ≈ 4 English characters ≈ 0.75 words (100 tokens ≈ 75 words).
2. A4 pages = round(words ÷ 500), taking a single-spaced A4 page of 11–12pt English prose as ≈ 500 words.
3. Lines of code = round(tokens ÷ 10). Source code tokenises more densely than prose, so a flat ~10-tokens-per-line planning heuristic is used and labelled as an approximation.

Ranking is a pure sort on the static dataset. The default view is context window descending, with ties broken deterministically by provider order then model name, so the order never shifts between renders. The search box does a case-insensitive substring match on the model name and provider, and the provider chips filter the set without touching the conversion maths.

Because both conversions come from the same base ratio, they are internally consistent: words computed directly (tokens × 0.75) equal words computed through characters (tokens × 4 characters ÷ 5.33 characters-per-word) to the last decimal. The module ships acrossCheckWords()function that proves this equality, so the figures are reproducible rather than asserted. The table also shows each model's max output tokens — a separate, usually much smaller cap — because a huge input window paired with a small output limit is a frequent and expensive surprise.

• OpenAI — Models & context length
• Anthropic — Models overview
• Google — Gemini API models
• Meta — Llama model cards
• DeepSeek — API documentation
• xAI — Grok models
• Mistral — Models overview
• OpenAI Help Center — What are tokens and how to count them (0.75 words/token ratio)

Every context-window and max-output figure was cross-checked against the provider docs above on 2026-07-03. The token-to-words, pages and code-line ratios are documented averages; use the AI Token Counter for an exact count of your own text.