ModelRefs / Logit Bias — AI Glossary
Logit Bias — AI Glossary
A per-token additive bias applied to logits before sampling, forcing or forbidding specific tokens from appearing. Supported by OpenAI and compatible APIs.
Overview
Logit bias maps token IDs to a float adjustment (–100 to +100). Setting –100 effectively bans a token; +100 forces it. Used to constrain output vocabulary for structured generation (JSON keys, binary yes/no), restrict safety-filtered words, or favor specific response styles. Supported by OpenAI and compatible APIs.
Reference details
| Topic | inference |
|---|---|
| Also known as | token bias |
| Last reviewed | 2026-06-24 |
Related terms
Continue your research
Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Logit Bias — AI Glossary.
Frequently asked questions
What is Logit Bias?
A per-token additive bias applied to logits before sampling, forcing or forbidding specific tokens from appearing.
Is Logit Bias the same as token bias?
Yes — token bias are common aliases for Logit Bias.
What concepts are related to Logit Bias?
Closely related concepts include logprobs, sampling, stop sequence.