A million tokens, measured in
PDFs are expensive to read properly: preserving layout, tables and figures costs ~1,500 tokens a page. A million tokens digests about 650 pages of dense reports.
1 block = 1 PDF page
≈ 650 PDF pages
visually-processed PDF pages
When a model reads a PDF page as both text and image (preserving layout, tables and figures), each page costs ~1,500 tokens. Text-only extraction is cheaper: ~2,000 pages per million tokens. For example, the IPCC AR6 Synthesis Report (81-page Longer Report, often cited as 85 pages) is about 121,500 tokens, so a million-token model could read it roughly 8 times over.
basis: ~1,500 tokens per page with layout
PDFs are the most token-intensive common format because they carry layout, fonts, tables, and figures alongside text. Processing a PDF as a rendered page costs roughly 3x more tokens than extracting its text alone. For enterprises running AI over contracts, research papers, or financial reports, this distinction directly affects architecture decisions and cost estimates.
Token counts for well-known documents, read with full layout (~1,500 tokens/page) or as extracted text (~500 tokens/page). Cost is to read the document once as input.
| Document | Pages | Tokens (layout) | Tokens (text only) | Fits in 1M | GPT-6 Luna | Claude Fable 5.1 |
|---|---|---|---|---|---|---|
| Bitcoin whitepaper | 9 | 13,500 | 4,500 | 74.1× | <$0.01 | $0.14 |
| "Attention Is All You Need" | 15 | 22,500 | 7,500 | 44.4× | <$0.01 | $0.22 |
| IPCC AR6 Synthesis Report (Longer Report) | 81 | 121,500 | 40,500 | 8.2× | $0.01 | $1.21 |
| GDPR (Official Journal) | 88 | 132,000 | 44,000 | 7.6× | $0.01 | $1.32 |
| GPT-4 Technical Report | 100 | 150,000 | 50,000 | 6.7× | $0.01 | $1.50 |
| Typical SEC 10-K filing | 200 | 300,000 | 100,000 | 3.3× | $0.03 | $3.00 |
| Mueller Report | 448 | 672,000 | 224,000 | 1.5× | $0.07 | $6.72 |
Page counts are for the published PDF editions; token counts are estimates and vary by tokenizer and page density.
Everything on this page is exactly one million tokens. So the price of having a model read all of it is simply each model's per-million rate. Updated 2026-10-02.
| Model | Read all of it (input) |
|---|---|
| GPT-6 Astra | $10 |
| Claude Fable 5.1 | $10 |
| Claude Opus 5 | $5 |
| Claude Opus 5.5 | $4 |
| GPT-6 Sol | $2 |
| Gemini 3.1 Pro | $2 |
| Claude Sonnet 5 | $2 |
| Claude Haiku 4.5 | $1 |
| Gemini 3.8 Flash | $0.75 |
| Gemini 3.5 Flash-Lite | $0.30 |
| GPT-6 Luna | $0.10 |
From $0.10 to $10 for the same million tokens. Try your own budget in the calculator →
Or start over with the full picture — scale, prices, and the budget calculator.