GPT-5 Token Counter

Count tokens for GPT-5, GPT-5 Mini, and GPT-5 Nano. Uses the real tiktoken WebAssembly with o200k_base — the exact tokenizer OpenAI bills with. Result matches your invoice within 1-2% for plain text (the residual is tool-call schema overhead, which billing includes but this counter doesn't model).

Advanced

How to use the GPT-5 Token Counter

Paste text, pick GPT-5 / Mini / Nano. The result is exact — billed-token-equivalent — for plain prompts. Cached input pricing (90% discount) kicks in when your system prompt is stable enough to hit OpenAI's prompt cache; toggle to see the discounted cost.

GPT-5 vs GPT-5 Mini vs GPT-5 Nano

Same o200k_base tokenizer across all three — token counts are identical. Only the per-token price differs: Mini is 5x cheaper than GPT-5, Nano is 25x cheaper. For workloads where Nano's quality is sufficient (routing, classification, short generation), the cost difference dominates the choice.

Reasoning behavior differs too. GPT-5 spends more on internal reasoning tokens (billed as output) for hard problems; Nano is lighter. If you're calling for a one-line answer to a simple question, Nano is almost always the right pick.