Claude token counter
Anthropic does not publish Claude's tokenizer, so browser-side tools can only estimate. This counter does both honestly: a live estimate as you type, and one click asks Anthropic's own count-tokens API, so the number you see is the number the API bills.
0 words · 0 characters
Claude models at a glance
Your text counted against every current Claude model. Counting against every provider at once? Use the full token counter, or see every model's limit in the context-window comparison.
| Model | Your text | Context window | Max output | Input $/M tokens |
|---|---|---|---|---|
| Claude Opus 4.8 no long-context surcharge | — | 1M ~750K words | 128K | $5 |
| Claude Sonnet 5 | — | 1M ~750K words | 128K | $3 |
| Claude Haiku 4.5 | — | 200K ~150K words | 64K | $1 |
Windows and prices verified against provider documentation on 2026-07-07.
Claude token questions, answered
How does Claude count tokens?
With Anthropic's own private tokenizer, which is not published for local use. Rules of thumb (about four characters or three quarters of a word per token) get close for English prose, but the only exact number comes from Anthropic's count-tokens API, which is what the exact-count button asks.
Is this count exact?
The live count while you type is an estimate. Click the exact-count button and the counter asks Anthropic's counting API directly; that result is exact for the listed Claude models.
How many tokens can Claude read at once?
Claude Opus 4.8 and Sonnet 5 take 1,000,000 tokens; Claude Haiku 4.5 takes 200,000. Note the claude.ai app caps consumer plans at 200K tokens regardless of model; the bigger windows apply over the API and in apps like NodePad that call the models directly.
Why does my Claude Pro limit run out faster in long chats?
Chat apps resend the whole conversation with every message, so late messages carry the entire thread with them. Counting your context, and trimming what the model does not need, is the practical way to stretch a usage-based limit.
Counting tokens is managing context by hand. NodePad is a canvas for doing it on purpose: fork side questions, keep every path, and choose exactly what Claude (or any other model) sees.
Try NodePad free