What is an AI token, and what does one cost?
An AI token is the unit a language model reads and writes, roughly four characters or three-quarters of an English word, and the unit its price is quoted in. As of September 18, 2026, a million input tokens costs from $0.030 to $150 across the 157 priced models in our registry; output tokens cost more, and cached input tokens less. A 500-word email is about 670 tokens: $0.00013 to read on GPT-5.6 Luna, $0.0067 on GPT-6 Astra.
A token is the unit the model reads, writes and bills
What it is
Before a model sees your text, a tokenizer cuts it into pieces from a fixed vocabulary: whole common words, fragments of longer words, numbers, punctuation and spaces. Each piece is a token. The model reads tokens in and produces tokens out, one at a time.
How big it is
In English prose a token is about four characters, or three-quarters of a word, so 1,000 words is roughly 1,300 tokens. Code, JSON, numbers and languages other than English take more tokens per word because they split into smaller pieces.
Why it is the unit of price
Providers charge per million tokens, with separate rates for input (what you send), cached input (a repeated prefix the provider does not recompute) and output (what the model writes). Output costs several times input because each output token is a full pass through the model.
How many tokens is that?
Counts on this page use OpenAI's o200k_base tokenizer, the one behind the current GPT models. Anthropic, Google and open-weight models use their own tokenizers, so the same text is a different number of tokens on each; the ratios are close for English prose and drift for code, other languages and JSON. Counts below were computed when this page was built; paste your own text into the token counter for counts per model.
| Text | Characters | Words | Tokens | Characters per token |
|---|---|---|---|---|
| A greetingHello, world! | 13 | 2 | 4 | 3.25 |
| A plain English sentenceThe quick brown fox jumps over the lazy dog. | 44 | 9 | 10 | 4.4 |
| A sentence about tokensTokenization splits text into pieces a model can price. | 55 | 9 | 11 | 5 |
| One long wordsupercalifragilisticexpialidocious | 34 | 1 | 10 | 3.4 |
| A JSON fragment{"ticket_id": "EX-012", "category": "feature"} | 46 | 4 | 15 | 3.07 |
| A short email (about 120 words)Hi Sam, thanks for sending the revised proposal over on Friday. I read it through twice over the weekend and I think the structure is much clearer now, especially the section on timelines. Two things before we send it to the client: the pricing table still shows the old rate for the second phase, and the appendix references a diagram that is not attached. Could you fix both by Wednesday? If you send me the updated file by then I will do a final read and get it out the door on Thursday morning. Let me know if anything on the timeline is unrealistic from your side. Best, Alex | 580 | 109 | 120 | 4.83 |
What a million tokens costs today
List rates per million tokens from the BenchLM pricing registry, updated September 18, 2026, for ten models across the price range. Across all 157 priced models the cheapest row is Qwen3.7 Flash at $0.030 input and $0.130 output, and the most expensive is o1-pro at $150 and $600. The full sortable table is on the LLM pricing page.
| Model | Input per 1M | Cached input per 1M | Output per 1M |
|---|---|---|---|
| GPT-5.6 LunaOpenAI | $0.200 | $0.020 | $1.20 |
| DeepSeek V4.1 FlashDeepSeek | $0.300 | $0.006 | $1.20 |
| Claude Haiku 4.5Anthropic | $1.00 | $0.100 | $5.00 |
| Gemini 3.5 FlashGoogle | $1.50 | $0.150 | $9.00 |
| GPT-5.6 TerraOpenAI | $2.00 | $0.200 | $12.00 |
| Claude Sonnet 5Anthropic | $2.00 | $0.200 | $10.00 |
| Grok 4.6xAI | $2.00 | $0.500 | $6.00 |
| GPT-5.6 SolOpenAI | $4.00 | $0.400 | $20.00 |
| Claude Opus 5Anthropic | $5.00 | $0.500 | $25.00 |
| GPT-6 AstraOpenAI | $10.00 | $1.00 | $50.00 |
What everyday units cost
Arithmetic on the list rates above, at about 0.75 words per token: the same email, report or book read as input or written as output on a budget, a mid-range and a frontier model. A million tokens is the unit prices are quoted in; a single token costs a fraction of a thousandth of a cent, and the bill comes from volume.
GPT-5.6 Luna
| Unit | Tokens | As input | As output |
|---|---|---|---|
| A 500-word email | 670 | $0.00013 | $0.00080 |
| A 10-page report (about 5,000 words) | 6,700 | $0.00134 | $0.00804 |
| A 300-page book (about 90,000 words) | 120,000 | $0.024 | $0.144 |
| A million tokens | 1,000,000 | $0.200 | $1.200 |
Claude Sonnet 5
| Unit | Tokens | As input | As output |
|---|---|---|---|
| A 500-word email | 670 | $0.00134 | $0.00670 |
| A 10-page report (about 5,000 words) | 6,700 | $0.013 | $0.067 |
| A 300-page book (about 90,000 words) | 120,000 | $0.240 | $1.200 |
| A million tokens | 1,000,000 | $2.000 | $10.000 |
GPT-6 Astra
| Unit | Tokens | As input | As output |
|---|---|---|---|
| A 500-word email | 670 | $0.00670 | $0.034 |
| A 10-page report (about 5,000 words) | 6,700 | $0.067 | $0.335 |
| A 300-page book (about 90,000 words) | 120,000 | $1.200 | $6.000 |
| A million tokens | 1,000,000 | $10.000 | $50.000 |
Put your own numbers in: the pricing calculator applies each model's input, cached and output rates to a monthly workload, and the token counter turns text into tokens per model first.
Keep these assumptions with the decision. The estimate above is today’s list rates against the workload you typed; both move. Save it as a Radar cost scenario and the next review starts from the same numbers, with the source of every rate attached.
Where token prices are heading
- Token Price Index, frontier tier (2026-09)
- 16
- 84% below the 2023-03 base of 100
- Median blended price
- $6.00
- per million tokens across 21 frontier constituents
Questions
What is a token in AI?
A token is the unit a language model reads and writes: a word, part of a word, a number or a punctuation mark, chosen by the model's tokenizer. In English prose one token is about four characters or three-quarters of a word. Prices are quoted per million tokens, separately for input, cached input and output.
How many tokens is a word?
About 1.3 tokens per English word on average, so 750 words is roughly 1,000 tokens. Short common words are one token; longer or rarer words split into several. On this page "supercalifragilisticexpialidocious" is 10 tokens on OpenAI's o200k_base tokenizer, while "Hello, world!" is 4.
How much do AI tokens cost?
As of September 18, 2026, a million input tokens costs from $0.030 to $150 across the 157 priced models in the BenchLM registry, and a million output tokens from $0.130 to $600. A single token therefore costs a fraction of a thousandth of a cent; the bill comes from volume, and from output tokens costing several times input tokens.
Why do output tokens cost more than input tokens?
Generating a token means running the whole model once per token, one after another, while reading input is processed in parallel. Providers price that difference: on the table above output is typically five to six times the input rate. Cached input is cheaper still, usually a tenth of the input rate, because a repeated prefix is not recomputed.
Are tokens the same across models?
No. Each provider has its own tokenizer, so the same text is a different number of tokens on GPT, Claude, Gemini or an open-weight model, and prices per million tokens are not directly comparable without that count. Our token counter shows the count per model; the pricing calculator applies each model's own rates to your token mix.