GPT Token Counter

OpenAI models use byte-pair encoding (BPE): text is split into chunks from a fixed vocabulary, and common words become single tokens while rare words break into pieces. o200k_base has a vocabulary of about 200,000 tokens.

tiktoken's model table maps GPT-4o and later models up to GPT-5 to o200k_base. At the time of writing, GPT-6 model names aren't in tiktoken's table yet (see openai/tiktoken issue #608), so for GPT-6 the o200k_base count is the closest published tokenizer, not a confirmed match.

Exact for o200k_base. The count comes from the tokenizer itself, running in your browser.

For GPT-6 models, the count is exact for o200k_base but OpenAI hasn't confirmed GPT-6 uses that encoding. Check the usage field on a real API response if you need the billed number.

Context windows and input prices from the pricing data used across these tools.

• OpenAI's rule of thumb for English is about 4 characters, or three-quarters of a word, per token. Other languages usually need more tokens per character.
• Leading spaces are part of tokens: " hello" and "hello" are different tokens.
• Chat requests add a few tokens of message framing per message, on top of your text.

Sources: OpenAI Help: What are tokens and how to count them? · OpenAI Cookbook: How to count tokens with tiktoken · openai/tiktoken issue #608: Update for gpt-6 models

• Claude token counter →
• Gemini token counter →
• Llama token counter →
• DeepSeek token counter →
• Mistral token counter →

Or compare every family at once on the main token counter.