OpenAI Token Counter

OpenAI prices every API call by the token — a sub-word unit their tokenizer splits your text into. The token count determines both what a request costs and whether it fits in the model's context window. This tool counts your text using o200k_base, the same tokenizer OpenAI uses for all current models, and shows the exact cost at that model's published rate.

The same text produces different token counts on different model families because each uses its own tokenizer vocabulary. However, all current OpenAI models (GPT-5.x, GPT-4.x, o-series) share the o200k_base encoding, so the token count is consistent across them — only the price changes.

It is for developers and prompt engineers who need to know exactly what a request will cost before calling the API: budgeting for batch jobs, checking whether a document fits, comparing model costs, or optimizing prompts for token efficiency. If you need to compare across vendors (OpenAI vs Claude vs Gemini), use the LLM Token Counter instead — this tool is OpenAI-only and uses a real encoding rather than an approximation.

The tool runs o200k_base, OpenAI's current tokenizer encoding, over your text. This is the same encoding used by GPT-5.6, GPT-5, GPT-4o, o3, and all other current OpenAI models — the count is exact, not an approximation.

Cost is calculated as: (token count ÷ 1,000,000) × model's per-million-token rate. The tool shows three pricing tiers: • Standard — normal API requests • Batch — 50% discount for non-urgent workloads processed within 24 hours • Flex — reduced rate for testing and development with lower priority

For models with long-context pricing (>128K tokens), the tool applies the appropriate rate based on your input length. Cached input pricing is also shown where applicable.