Prompt Engineering for Each Model

Different models respond better to slightly different prompting styles. You do not need a completely different prompt for every model, but it helps to match the prompt to the model's role in your workflow.

Claude Sonnet 5.5

Best for writing, analysis, and code.

• give it context
• ask for structured outputs when the task is complex
• use it when nuance matters

GPT-6.1 Sol and GPT-6 Luna

Compare Luna and Sol on your own task. Luna has lower standard token rates.

• use direct, specific instructions
• include examples when output format matters
• verify results for simpler tasks before using them

Claude Haiku 5.5

Good for structured tasks, long documents, and faster lower-cost work.

• ask for tables, JSON, or structured summaries when useful
• check sources and document contents before relying on analysis
• compare output against a representative prompt

Claude Opus 5.5

Use when the task is important enough to justify a premium model.

• provide rich context
• ask for detailed reasoning or careful review
• save it for higher-value tasks

Current search-enabled chat models

Use when the task depends on live information from the web.

• ask factual questions
• request sources when you need them
• verify original sources for broader investigations

GLM-5.3

Use when you want a lower-cost reasoning-oriented model.

• present the problem clearly
• ask for step-by-step reasoning
• keep it focused on analytical tasks

Grok 4.7

Use as another general-purpose option when you want to compare outputs across providers.

• use direct prompts
• compare its answer against Claude or GPT when the task matters

General advice

• give the model the minimum useful context, not every detail you know
• ask for the output format you want
• switch models when the task changes instead of trying to make one model do everything
• use lower-cost models for drafts and premium models for final review