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

Best for writing, analysis, and code.

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

GPT-5.5 and GPT-5.4 Mini

Best for broad general work. Mini is the cheaper option for lighter tasks.

• use direct, specific instructions
• include examples when output format matters
• keep Mini focused on simpler tasks

Gemini 3.1 Pro and Gemini 3 Flash

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

• ask for tables, JSON, or structured summaries when useful
• use Pro for heavier analysis
• use Flash for quick iteration

Claude Opus 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

Perplexity Reasoning and Deep Research

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

• ask factual questions
• request sources when you need them
• use Deep Research for broader investigations, not casual chat

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 4.5

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