OpenAI Token Counter -- GPT-5, GPT-4o Tokens

OpenAI Token Counter

OpenAI Token Counter estimates how many tokens your text will use with GPT-5.x, GPT-4o, GPT-4.1, and other OpenAI models. All current OpenAI models use the o200k_base encoding (~4 characters per token for English).

OpenAI Model Comparison

All current OpenAI models use the o200k_base encoding with a vocabulary of approximately 200,000 tokens.

*GPT-5.3-Codex released Feb 5, 2026 -- API pricing not yet announced; currently ChatGPT-only.

How OpenAI Tokenization Works

OpenAI models use Byte Pair Encoding (BPE) to split text into tokens. The current tokenizer is called o200k_base, which has a vocabulary of approximately 200,000 tokens. This is the same encoding used by all GPT-5.x, GPT-4o, and GPT-4.1 models.

What is BPE?

Byte Pair Encoding starts with individual bytes and iteratively merges the most frequent pairs into new tokens. This creates a vocabulary where common words like "the" or "and" are single tokens, while rare words get split into subword pieces.

The tiktoken Library

OpenAI provides an open-source Python library called tiktoken for exact token counting. To use it:

import tiktoken

enc = tiktoken.encoding_for_model("gpt-4o")
tokens = enc.encode("Hello, world!")
print(len(tokens))  # 4

Token Estimation Rules

• English text: ~4 characters per token (1 word = ~1.3 tokens)
• Code: ~3-3.5 characters per token (symbols split into separate tokens)
• CJK text: ~1.5-2 characters per token (each character may be multiple tokens)
• Numbers: each digit or small group is typically 1 token

Frequently Asked Questions

How many tokens does GPT-5 use per word?

GPT-5 uses the o200k_base tokenizer, which averages about 1.3 tokens per English word (or roughly 4 characters per token). Technical terms and code use more tokens per word. Non-English languages, especially CJK, use significantly more tokens.

What encoding does GPT-4o use?

GPT-4o uses the o200k_base encoding with a vocabulary of approximately 200,000 tokens. This is the same encoding used across all current OpenAI models including GPT-5.x and GPT-4.1.

How to count tokens for the OpenAI API?

You can use this free online counter for estimates. For exact counts, use the official tiktoken Python library: pip install tiktoken, then use tiktoken.encoding_for_model('gpt-4o') to get the encoder.

What is tiktoken?

tiktoken is OpenAI's open-source tokenizer library for Python. It provides exact token counts for all OpenAI models. It supports multiple encodings including o200k_base (current models) and cl100k_base (legacy models like GPT-4 Turbo).

What is the difference between GPT-5 and GPT-5.1?

GPT-5.1 is an updated version of GPT-5 with improved reasoning and instruction following. Both use the same o200k_base tokenizer and 400K context window, and have the same input pricing ($1.25/1M tokens).

How much does GPT-5 cost per 1000 tokens?

GPT-5 costs $0.00125 per 1K input tokens ($1.25 per 1M) and $0.01 per 1K output tokens ($10.00 per 1M). GPT-5 Mini is significantly cheaper at $0.00025 per 1K input tokens. Output tokens always cost more than input tokens.

Token Counters by Provider