GPT-6 Astra
OpenAI shows few-shot examples in a developer message so a model can infer the desired classification or response pattern from demonstrations.
AI Prompt
A reference directory for comparing practical prompt-design marks across current model families. It organizes reusable prompt patterns, chaining methods, evaluation habits, and failure checks around official model documentation so readers can distinguish a template pattern from a tested workflow or a known anti-pattern.
Models and OEM references
| Model | OEM reference | Part number | Price | Fact | Source |
|---|---|---|---|---|---|
| GPT-6 Astra | Structured Outputs with JSON Schema | OpenAI documents Structured Outputs as a way to constrain supported model responses to a supplied JSON Schema instead of relying only on free-form formatting instructions. | https://developers.openai.com/api/docs/guides/structured-outputs | ||
| GPT-5.5 | Few-shot input/output examples | OpenAI's prompt-engineering guide describes few-shot learning as steering a task with a small set of representative input and output examples. | https://developers.openai.com/api/docs/guides/prompt-engineering | ||
| Claude Opus 5 | XML-tag prompt structure | Anthropic recommends descriptive XML tags to separate instructions, context, examples, and other prompt sections when complex prompts need clearer structure. | https://docs.anthropic.com/en/docs/build-with-claude/prompt-engineering/prompt-templates-and-variables | ||
| Gemini 2.0 Flash | Vertex AI system instructions | Google documents system instructions as model-level guidance for behavior, persona, context, and response formatting before user prompts are processed. | https://cloud.google.com/vertex-ai/generative-ai/docs/learn/prompts/system-instruction-introduction |
Entries
Claude Opus 5
Anthropic describes explicit prompt chaining as useful when intermediate results must be inspected, including a self-correction pattern of draft, review against criteria, and refinement.
GPT-6 Astra
OpenAI describes an eval workflow as defining the task, running it on test inputs, analyzing results, and iterating on the prompt.
GPT-6 Astra
OpenAI lists vibe-based evaluation, overly generic metrics, biased eval datasets, and ignoring human feedback among evaluation anti-patterns.