- Abliteration
- Removing a model's refusals by projecting the direction that triggers them out of its weights. Arditi et al., 2024.
- Abuse monitoring
- Keeping prompts and outputs for a fixed window so a provider can investigate misuse later. More
- Application state
- Data an API keeps between requests, such as stored responses and files. More
- Batch API
- An endpoint that runs requests later for less, storing your inputs until then. More
- Context window
- How many tokens a model reads in one request, prompt and history included.
- Covered Models
- Anthropic models, such as Claude Fable 5.2, whose prompts it keeps 30 days. More
- Data processing addendum
- A contract that sets how a provider handles personal data for you. More
- Default retention
- How long a provider keeps prompts if you change nothing. More
- Flagged content
- Requests a classifier marks as likely policy violations, often kept longer. More
- Grounding
- Letting a model search the web or a map service while it answers. More
- Human review
- People at the provider reading your prompts. More
- Modified Abuse Monitoring
- An OpenAI control that keeps your content out of abuse logs, by approval. More
- Opt-out
- A setting that stops a provider training on your data. More
- Private Safety Processing
- An OpenAI option that writes safety records to the customer's own storage. More
- Prompt caching
- Reusing a processed prompt prefix to save cost, for minutes to a day.
- Retention period
- The time between a provider receiving your prompt and deleting it. More
- Stateless API
- An endpoint that keeps nothing between calls, so you send the full conversation each time. More
- Store parameter
- A request field that tells an API whether to keep the response. More
- Training on customer data
- Using your prompts and outputs to train or improve models. More
- Unpaid Services
- Google's term for free use of the Gemini API and AI Studio. More
- Usage logs
- Records of when you called an API and how much you used. More
- Zero data retention
- A provider not storing prompts or outputs after it returns the response. More
A
B
C
D
F
G
H
M
O
P
R
S
T
U
Z