AI token counter

A token is the unit a model reads and bills by: often a whole short word, sometimes part of a long one, a space or a punctuation mark. Each provider splits text differently, so the same prompt has a different count, and a different price, on every model.

Exact counts come from OpenAI's published o200k_base tokenizer, running in your browser. Estimates are marked with ≈ wherever a provider keeps its tokenizer private. Estimates are usually within about 10% for English prose and drift further for code and other languages.

GPT-4o, GPT-4.1, the o-series and the GPT-5 generation all use o200k_base, so the count you see is the count the API bills for plain text. Chat requests add a few tokens per message for roles and formatting. GPT-6 and GPT-5.6 are counted with the same tokenizer and marked as estimates until OpenAI confirms it.

Anthropic says Claude 4.7 and later models use a newer tokenizer that produces about 30% more tokens for the same text. That is why Claude Opus 5.5 can cost more per prompt than its per-token price suggests. This counter estimates older Claude models at 1.1 times the o200k count, and Claude 4.7 and newer about 30% above that.

For a bill-accurate figure, send the text to Anthropic's token counting endpoint with the model ID. It costs nothing to call. Claude Haiku 5.5 switches to higher prices once a prompt passes 100,000 tokens, and the counter applies that automatically.

Google gives a rule of thumb of about 4 characters per token, which this counter uses for every Gemini model. Gemini 4 Argon and Gemini 3.8 Flash both have price rises already announced on Google's pricing page. They are listed under each model below.

Claude counts one token per 28×28 pixel patch. Claude 4.7 and later accept up to 2,576px on the long edge and 4,784 tokens per image. Older models downsize to 1,568px and 1,568 tokens. A 1920×1080 screenshot is 2,691 tokens on new models and 1,560 on old ones.

Gemini counts 258 tokens for an image up to 384px on both sides. Larger images are cut into 768×768 tiles of 258 tokens each, so the same screenshot is 1,548 tokens. Use the Image tab to read the size of your own file without uploading it.

Gemini is the API family here that takes video files directly. It samples one frame per second at 263 tokens per second by default, about 100 at the low setting and 300 at high. The soundtrack adds 32 tokens per second, and audio-only files count at the same rate. A one-minute clip with sound comes to about 17,700 tokens.

Video generators such as Veo, Sora and Runway do not bill by input tokens. They charge per second of generated video, so a token counter cannot price them.