How much does GPT-3.5 Turbo cost per million tokens?

$0.50 per million input tokens and $1.50 per million output tokens.

What is the context window of GPT-3.5 Turbo?

16,385 tokens, covering input and output combined.

Which tokenizer does GPT-3.5 Turbo use?

GPT-3.5 Turbo uses the cl100k_base encoding. Counts are computed in your browser with the same tokenizer OpenAI publishes, so they are exact.

GPT-3.5 Turbo pricing and token costs

GPT-3.5 Turbo is a OpenAI model with a 16,385-token context window. Rates below are per million tokens, and are the ones in effect today.

Retiring. OpenAI removes this model on 2026-10-23. The named replacement is GPT-5.6 Terra.

In practice

Its 16,385-token window is by far the smallest here — around one sixty-fourth of the million-token models — and that is the number that decides whether it is usable at all, before price enters the conversation. Output is three times input, and it retires on 23 October 2026 with GPT-5.6 Terra named as the replacement. Like GPT-4 Turbo it uses cl100k_base rather than o200k_base, so paste the same text into both generations and the totals will not agree.

Pricing

Input, per 1M tokens: $0.50
Output, per 1M tokens: $1.50
Context window: 16,385 tokens
Tokenizer: cl100k_base

Pricing data last verified: 31 Aug 2026.

Provider pricing can change. Check current rates with OpenAI, Anthropic, and Google before making billing decisions.

Count your own prompt Compare all 15 models

Alternatives

Head to head

GPT-3.5 Turbo vs GPT-5.6 Luna

Articles that quote these figures

Common questions

How much does GPT-3.5 Turbo cost per million tokens?: $0.50 per million input tokens and $1.50 per million output tokens.
What is the context window of GPT-3.5 Turbo?: 16,385 tokens, covering input and output combined.
Which tokenizer does GPT-3.5 Turbo use?: GPT-3.5 Turbo uses the cl100k_base encoding. Counts are computed in your browser with the same tokenizer OpenAI publishes, so they are exact.
Is GPT-3.5 Turbo being retired?: OpenAI removes this model on 2026-10-23. The named replacement is GPT-5.6 Terra.

Figures come from OpenAI’s published list prices and are checked against every article on this site that quotes them at build time. They are still estimates of what a request costs — verify against your own billing before committing to them. See the methodology page for how these numbers are actually calculated.