Guides
Prompt engineering guides by model
Every model family reads prompts differently. Claude wants XML tags, GPT-6 Astra wants markdown sections, Gemini wants constraints last. These guides cover the structure each one responds to, with worked examples and links to the vendor documentation.
Anthropic
- Claude Opus 5 XML-tagged sections A practical Claude Opus 5 prompting guide: XML-tagged structure, controlling verbosity and scope, effort levels, and what to remove from older Claude prompts. Read the guide
- Claude Sonnet 4.5 XML-tagged sections Claude Sonnet 4.5 prompting guide: XML-tagged structure, stating the full task at once, bounding scope, and one skeleton for code, writing and analysis. Read the guide
- Claude Haiku 4.5 Compact XML-tagged sections Claude Haiku 4.5 prompting guide: short XML-tagged briefs, one deliverable per prompt, numeric length caps, and why Opus-style prompts backfire on it. Read the guide
OpenAI
- Gemini 3 Pro Context → task → constraints last Gemini 3 Pro prompting guide: context first and constraints last, positive grounding instead of "do not infer", split-step verification and personas. Read the guide
- Gemini 3 Flash Dense labeled blocks, constraints last Gemini 3 Flash prompting guide: dense labeled blocks, constraints as the final block, positive grounding, and the long-brief habits that hurt it. Read the guide