What Are the Official Maximum Output Token Limits for Various GPT Models?

Below is a summary based on the officially published tables on the models page.

GPT-3.5‑16k Models

Even though the GPT‑3.5‑16k variants support a larger context (about 16,385 tokens), the maximum output length remains capped at 4,096 tokens. This is evidenced by the table entry for the GPT‑3.5‑Turbo‑0125 variant.

GPT‑4 Models

For the original GPT‑4 models (as represented by variants such as gpt-4, gpt-4-0613, and the legacy gpt-4-0314), the specification shows a context window of 8,192 tokens and a corresponding maximum output of 8,192 tokens. Note that the GPT‑4‑Turbo variants are different; they are limited to a maximum output of 4,096 tokens even though their context window is much larger (128,000 tokens).

Official Documentation

These limits are officially documented in the model specification tables on the OpenAI platform documentation (https://platform.openai.com/docs/models). The relevant tables list the context window and maximum output tokens for each model variant.

So, to directly answer your question:

• GPT-3.5‑16k models: Maximum response length is 4,096 tokens.
• GPT‑4 (non-Turbo models): Maximum response length is 8,192 tokens.

For the most up-to-date detailed specifications, it’s best to refer directly to the model tables on the official documentation page.