GPT-3.5 Token Counter

Count tokens for GPT-3.5 Turbo. Uses the real tiktoken WebAssembly with cl100k_base — the tokenizer GPT-3.5 and the original GPT-4 share. For migrating away from GPT-3.5, compare against the GPT-5 token counter or GPT-4o counter on the same text — newer models use o200k_base which is typically 5-15% more efficient.

Advanced

How to use the GPT-3.5 Token Counter

Paste text. GPT-3.5 uses cl100k_base — a ~100K-token vocabulary. For the same English text, this tokenizer typically produces 5-15% more tokens than o200k_base (GPT-4o / GPT-5). For non-English, the gap can be larger.

Should you still use GPT-3.5?

By 2026, GPT-3.5 is rarely the right choice. GPT-4o Mini at $0.15 / $0.60 per million tokens is cheaper than GPT-3.5 ($0.50 / $1.50) AND substantially better. GPT-5 Nano at $0.05 / $0.40 is cheaper still. The only case for keeping GPT-3.5 is a long-running production workload where the output style is baked into downstream parsers and a model swap is more work than it's worth.

Frequently asked questions

Which tokenizer does GPT-3.5 use?
cl100k_base, the same tokenizer the original GPT-4 used. This tool runs the real tiktoken in WebAssembly, so counts are exact for plain text.

Will my count match newer GPT models?
No. GPT-4o and GPT-5 use o200k_base, which usually produces 5 to 15% fewer tokens on the same text. Use the matching counter for those models.

Is my text sent to OpenAI?
No. Tokenization happens locally in your browser, so your text never leaves your machine.
Embed this tool on your site

Free to embed, no attribution required (but appreciated). Paste this where you want the tool to appear: