ChatGPT Claude Gemini Token Counter | API Cost Estimator (July 2026 edition)
ChatGPT, Claude, Gemini. Token count, API cost, context usage — instant.
Data is never sent to the server. Everything runs locally in your browser. Safe for sensitive information.
About this tool
Paste text for GPT-5.6 / Claude Sonnet 5 / Fable 5 / Gemini 3.6 token counts, input/output costs, and context usage. OpenAI uses tiktoken exact.
Tool interface
Runs in your browser — nothing is uploaded Estimate
Cost estimate
Monthly uses GPT-5.6 Luna pricing ($1/$6 per 1M).
vs ChatGPT Web limit
Separate from API limits. Plus Instant ≈ 32K.
System prompt (optional)
API billing counts system + user together as input tokens.
| Model | Tokens | Usage | Input | Output |
|---|
OpenAI uses tiktoken (o200k_base) — same as the API. Claude / Gemini / DeepSeek counts are approximate (±8%). Share links store data in the URL hash only. Pricing from official pages (as of 2026-07-21).
How token counting works
Paste once to see character stats, OpenAI exact tokens (tiktoken o200k), and script-weighted approx counts for Claude / Gemini / DeepSeek with input/output cost and context usage %. Pricing syncs with our rate table.
ChatGPT Web vs API limits
ChatGPT Plus Instant is roughly 32K tokens; Thinking mode is closer to 256K. OpenAI API models (GPT-5.6, etc.) expose 1M+ context. The Web limit bar compares your prompt against app limits — not API limits.
Exact vs approx
OpenAI models use the same o200k_base tokenizer as the API (exact).
Claude and Gemini use proprietary tokenizers, so we apply measured CJK / Latin / other tok/char rates and label them approx (±8% typical).
Japanese uses more tokens than English
| Model | tok / char | Note |
|---|---|---|
| Gemini 3.6 Flash | ~0.57 | Efficient on Japanese |
| GPT-5.6 / o200k | ~0.68 | OpenAI exact |
| Claude Sonnet 5 | ~0.87 | approx |
System prompt & RAG chunks
Billing counts system + user together. Add RAG snippets to the main textarea or system field before trusting the numbers. Gemini 3.1 Pro switches to long-context tier pricing above 200K input tokens (shown as a badge in the table). GPT-5.6 tiers switch above 272K input.
RAG chunk reference (tiktoken measured / GPT-5.6 1M ctx)
Values measured with o200k_base (2026-07-19).
| Chunk | Chars | Tokens | 1M ctx | 32K Web |
|---|---|---|---|---|
| Short FAQ (Japanese) | 133 | 68 | 0.01% | 0.21% |
| Tech article section | 1,763 | 970 | 0.1% | 3.03% |
| ~10 PDF pages | 7,542 | 3,972 | 0.4% | 12.41% |
| Full internal wiki (risk) | 80,000 | 42,779 | 4.28% | 133.68% |
Pricing sources
Related
Usage
- Paste your prompt (autofocus on load)
- Check OpenAI exact tokens plus all-model costs and usage %
- If usage is tight or over, shorten or switch models
When to use
Examples
Tips
- OpenAI models use exact tiktoken (o200k). Claude/Gemini use script-weighted approx (±8%).
- ChatGPT Web ≈ 32K vs API 1M+ — use the Web limit bar separately.
- System prompt and RAG chunks count toward input tokens.
FAQ
How accurate is the count?
What is a token?
Web vs API limits?
Does system prompt count?
Is ChatGPT Claude Gemini Token Counter | API Cost Estimator (July 2026 edition) free?
Is data sent to a server?
Supported browsers?
Offline use?
vs CLI or desktop apps?
When to use it?
Usage example?
Main features?
How is this different from similar tools?
Search keywords
Basic workflow
- Paste your prompt (autofocus on load)
- Check OpenAI exact tokens plus all-model costs and usage %
- If usage is tight or over, shorten or switch models
Practical use cases
- API cost estimates, prompt optimization, context window checks, ChatGPT long-thread length checks.
- Paste text for GPT-5.6 / Claude Sonnet 5 / Fable 5 / Gemini 3.6 token counts, input/output costs, and context usage. OpenAI uses tiktoken exact.
- OpenAI models use exact tiktoken (o200k). Claude/Gemini use script-weighted approx (±8%).
- ChatGPT Web ≈ 32K vs API 1M+ — use the Web limit bar separately.
- System prompt and RAG chunks count toward input tokens.
- Instant token count
- OpenAI tiktoken exact
Privacy & data handling
Input is processed locally in your browser—nothing is sent to our servers.
Things to watch out for
- Treat ChatGPT Claude Gemini Token Counter | API Cost Estimator (July 2026 edition) as a quick check — re-verify critical values in your editor or CI before shipping.
- This tool runs locally: closing the tab clears unsaved input. Copy results you need to keep.
- Very large pastes can freeze a tab briefly — wait for the result before closing the tab.
- OpenAI models use exact tiktoken (o200k). Claude/Gemini use script-weighted approx (±8%).
- ChatGPT Web ≈ 32K vs API 1M+ — use the Web limit bar separately.