Paste a document below. This estimates its token count and marks which tools can hold it in a single conversation.
Estimated tokens: 0
| Tool | Published limit | Your document |
|---|---|---|
| ChatGPT | 1,050,000 tokens on GPT-6 Astra (922K max input, 128K max output); GPT-6.1 Sol inside ChatGPT Work defaults to 500K (checked 2026-10-29) | — |
| Claude | Up to 1M tokens, varying with the selected model (checked 2026-10-29) | — |
| Gemini | 1M tokens (Gemini 4 Argon unified context; Pro models take 1M input and 64K output) (checked 2026-10-29) | — |
| Grok | 500K tokens (grok-4.7) (checked 2026-10-29) | — |
| DeepSeek | 1M tokens of context; up to 384K tokens of output (checked 2026-10-29) | — |
| Doubao | 1M tokens (doubao-seed-2.1-pro; Volcano Ark prices 0-1024K input in bands) (checked 2026-10-29) | — |
| Kimi | 1,048,576 tokens (K3 series); 262,144 tokens on the K2 series (checked 2026-10-29) | — |
| Qwen | 1,000,000 tokens (qwen3.8-max; 991,808 max input, 131,072 max output) (checked 2026-10-29) | — |
| Yuanbao | not published | — |
| GitHub Copilot | not published | — |
| Perplexity | not published | — |
Token estimates are approximate: the real count depends on the tokenizer each vendor uses, and on whether you attach files (attachments get processed differently from pasted text). Treat this as a pre-check that saves you from pasting a 120-page PDF into a 200K window and wondering why the ending was ignored.
Two practical rules: leave 30–40% headroom, because the model's reply and your follow-up questions also occupy the window; and if a document sits just under the limit, split it by chapter and ask one question per section — long inputs that technically fit still degrade recall near the edges.