GPT-4o Token Counter

Count tokens for GPT-4o and GPT-4o Mini. Uses the real tiktoken WebAssembly with o200k_base. Token counts match OpenAI's billing exactly for plain text. Also handles image and audio input rough estimates if you toggle them.

How to use the GPT-4o Token Counter

Paste text, pick GPT-4o or GPT-4o Mini. Same o200k_base tokenizer as GPT-5 — token counts identical, only price differs. The cached-input discount on GPT-4o is 50% (vs 90% on GPT-5), reflecting OpenAI's tiered cache pricing.

When GPT-4o still makes sense in 2026

GPT-5 launched in August 2025 and is now the default for most workloads. GPT-4o remains relevant when you need image / audio input (4o's native multimodal is mature; GPT-5 inherits it but some specific endpoints still default to 4o), when you've cached a large prompt that you don't want to re-cache against a new model, or when GPT-5's reasoning behavior changes output structure in ways that break existing parsers.