ChatGPT Token Counter
Paste any text and see exactly how many tokens ChatGPT will bill you for — counted in your browser with OpenAI’s own o200k_base vocabulary, not estimated from character counts. All 11 GPT models are priced below, from GPT-5.6 Sol at $4.00 per 1M input tokens down to GPT-4o Mini at $0.150.
- Countingo200k_base
- Context128K–1M
- Input$2.00
- Output$12.00
- Cached$0.20010% of input
Count ChatGPT tokens
Opens on GPT-5.6 Terra. Switching models re-prices the same text but never changes the count — every ChatGPT model in the list shares the o200k_base vocabulary.
Drop files here or click to browse
Supports PDF, TXT, MD, JSON, CSV, XML — max 10MB
Count as an API request
Providers also bill the role and boundary tokens that wrap your text. Anthropic's own example — a short system prompt plus one message — costs 14 input tokens for 7 tokens of visible text.
Are these ChatGPT token counts exact?
Yes, for every model on this page. OpenAI publishes the o200k_base vocabulary its models use, and this page loads that vocabulary and runs the same byte-pair merges the API runs. The number you see is the number you are billed for, not an approximation. It all happens in your browser, so nothing you paste is uploaded anywhere.
ChatGPT is the exception rather than the rule here. Of the 56 models priced on this site, OpenAI’s are the only ones counted exactly — every other provider keeps its tokenizer private, so those counters estimate. One caveat that applies even to an exact count: a chat request is more than the text you send. Role markers, the system prompt, tool and function schemas, and any attached files all consume tokens from the same window, so a real API call always bills slightly above the figure below.
ChatGPT models and prices
11 OpenAI models, all counted exactly. Output costs four to eight times more than input on every row — GPT-5.6 Sol is $4.00 in and $20.00 out, GPT-4o Mini is $0.150 and $0.600 — so the shape of your traffic decides which model is cheapest for you, not the tier name.
| Model | Inputper 1M tokens | Cachedper 1M tokens | Outputper 1M tokens | Context | Counting | Your cost |
|---|---|---|---|---|---|---|
| GPT-5.6 Sol | $4.00 | $0.400 | $20.00 | 1M | exact | — |
| GPT-5.6 Terra | $2.00 | $0.200 | $12.00 | 1M | exact | — |
| GPT-5.6 Luna | $0.200 | $0.020 | $1.20 | 1M | exact | — |
| GPT-5.5 | $5.00 | $0.500 | $30.00 | 1M | exact | — |
| GPT-5.4 | $2.50 | $0.250 | $15.00 | 1M | exact | — |
| GPT-5.4 mini | $0.750 | $0.075 | $4.50 | 400K | exact | — |
| GPT-5.4 nano | $0.200 | $0.020 | $1.25 | 400K | exact | — |
| GPT-5.1 | $1.25 | $0.125 | $10.00 | 400K | exact | — |
| GPT-4.1 | $2.00 | $0.500 | $8.00 | 1M | exact | — |
| GPT-4o | $2.50 | $1.25 | $10.00 | 128K | exact | — |
| GPT-4o Mini | $0.150 | $0.075 | $0.600 | 128K | exact | — |
Rates in USD per 1M tokens, read from OpenAI’s own pricing page on August 22, 2026. Providers change prices without notice.
ChatGPT’s context window
GPT-5.6 Terra and GPT-5.6 Sol accept 1,050,000 tokens in a single request — roughly 1,567 pages of ordinary prose, at about 670 tokens a page. That window covers the whole request, not just your prompt: system instructions, conversation history, attached documents and the answer being written all draw from it.
Not every ChatGPT model has that much room. GPT-5.1 and GPT-5.4 mini hold 400K, and the GPT-4o pair stops at 128K — 191 pages — so a prompt GPT-5.6 Sol accepts happily is rejected outright by GPT-4o Mini. Two things bite at the top of the range: OpenAI bills prompts above 272,000 input tokens at a higher rate than the one shown, and cost scales linearly with length, so sending 500,000 tokens of context to GPT-5.6 Terra costs $1.00 before the model writes a word.
What ChatGPT tokens cost
Three request shapes priced on GPT-5.6 Terra, GPT-5.6 Sol and GPT-4o Mini. Input and output bill at different rates, so the ratio between them matters as much as the total.
| Request | GPT-5.6 Terra | GPT-5.6 Sol | GPT-4o Mini |
|---|---|---|---|
| One chat turn1,000 in / 500 out | $0.0080 | $0.014 | $0.0004 |
| Summarising a report100,000 in / 2,000 out | $0.224 | $0.440 | $0.016 |
| A batch of a thousand1,000,000 in / 50,000 out | $2.60 | $5.00 | $0.180 |
Cached input moves these numbers more than the model choice does. Every GPT-5 model on this page reads cached input at exactly 10% of its standard rate — $0.200 instead of $2.00 on GPT-5.6 Terra — so a system prompt reused across a conversation is nearly free after the first turn. The older models are less generous: GPT-4.1 reads cache at 25% of input, and both GPT-4o and GPT-4o Mini at 50%. Caching only matches an identical prefix, so put the stable parts of a prompt first.
ChatGPT token counter FAQ
How many tokens is 1,000 words in ChatGPT?
About 1,330 tokens for ordinary English prose — roughly 1.33 tokens per word. In the other direction, 1,000 tokens is about 752 words. The ratio is an average, not a rule: common English words are usually one token each, while code, JSON, URLs, emoji and non-Latin scripts split into several. That is exactly why the counter above runs the real tokenizer instead of multiplying — paste your own text and the number is exact.
Does ChatGPT use the same tokenizer as the OpenAI API?
Yes. GPT-4o and every model released after it use the o200k_base vocabulary, which is what this page loads, so a given string produces the same count in ChatGPT and through the API. What differs is the wrapping: a chat request adds role markers, a system prompt and any tool schemas, and all of that is billed. Count your text here to size it, then expect a real API call to report a slightly higher number.
What is GPT-5.6’s context window?
1,050,000 tokens on GPT-5.6 Terra and GPT-5.6 Sol — about 1,567 pages of prose. GPT-5.1 and GPT-5.4 mini hold 400K, and GPT-4o and GPT-4o Mini hold 128K. Prompt and completion share the window, so a request that fills it leaves no room for an answer — reserve whatever you set as the maximum output length.
How much does 1M tokens cost on ChatGPT?
$2.00 to send 1M input tokens to GPT-5.6 Terra and $12.00 to receive 1M output tokens. GPT-5.6 Sol costs $4.00 and $20.00; GPT-4o Mini costs $0.150 and $0.600. Real traffic is lopsided — you send far more than you receive — so a 100,000-token document summarised in 2,000 tokens costs $0.224 on GPT-5.6 Terra, not the headline per-million figure.
What does cached input pricing actually save?
Up to 90% on the repeated part of a prompt. Every GPT-5 model here reads cached input at exactly 10% of its standard rate: $0.200 against $2.00 on GPT-5.6 Terra. Older models discount less — 25% on GPT-4.1, 50% on GPT-4o and GPT-4o Mini. The cache matches on an identical prefix only, so it rewards a fixed system prompt and punishes one that interpolates a timestamp.
Is this the same as tiktoken?
It uses the same vocabulary. tiktoken is OpenAI’s tokenizer library; o200k_base is the encoding it loads for GPT-4o and newer, and this page runs those same merge ranks locally, so counts match tiktoken for plain text. The o200k_harmony encoding used by OpenAI’s open-weight releases reuses these ranks and only adds control tokens, so ordinary prose counts identically there too.
Counting for a different model?
The same paragraph tokenizes differently on Claude, Gemini, Grok, DeepSeek and Qwen, and it prices very differently — output rates across the full catalogue span more than two orders of magnitude. Compare 56 models from 11 providers side by side, or jump straight to another brand’s counter below.
New to tokens? Read how tokenization works