What is Calibration?
Calibration is an evaluation concept used to measure language model behaviour. It is part of the vocabulary teams use when they compare models, design prompts or review an AI system.
Definition
Calibration is an evaluation concept used to measure language model behaviour. It is part of the vocabulary teams use when they compare models, design prompts or review an AI system. The term is used in evaluation discussions, where precision matters: two teams can use the same word to mean slightly different things. In this glossary, calibration is treated as a practical concept rather than a formal specification.
Why it matters
Evaluation terms cover how capability, reliability and safety are measured. They matter because a benchmark score is only useful if you understand what the test measures and where it can mislead.
Example
Suppose you are comparing Pythia 8B with another model for a production task. Calibration is one of the factors behind the numbers you see: Pythia 8B lists a 32,768-token context window, costs $0.16 per million input tokens and scores 49.7 on TruthfulQA. Understanding the term helps you decide which of those differences actually matter for your workload.
How to evaluate it
- Check whether calibration is defined the same way in the documentation you are reading.
- Ask which measurement or trade-off the term is describing before comparing numbers.
- Separate vendor claims from independently observed behaviour on your own data.
- Write down the definition your team uses so reviews and evaluations stay consistent.
Related terms
Related guides
FAQ
What does Calibration mean?
Calibration is an evaluation concept used to measure language model behaviour. It is part of the vocabulary teams use when they compare models, design prompts or review an AI system.
Why does Calibration matter?
Evaluation terms cover how capability, reliability and safety are measured. They matter because a benchmark score is only useful if you understand what the test measures and where it can mislead.
How is Calibration used in practice?
Suppose you are comparing Pythia 8B with another model for a production task. Calibration is one of the factors behind the numbers you see: Pythia 8B lists a 32,768-token context window, costs $0.16 per million input tokens and scores 49.7 on TruthfulQA. Understanding the term helps you decide which of those differences actually matter for your workload.