It has been such a Long time since I’ve written a blog post but there’s so much going on the world right now I thought it might be useful only for myself but others as well.
Today wanna talk about AI models and whole idea of best practices? These models produce different companies so only makes sense for there to be nuances between them plus AI is evolving at such a rapid rate. He cannot honestly expect something that well a year ago to work well today.
I’ve been trying to keep up with all source of influencers and newsletters on the essential of good prompting and it is near possible so here’s my attempt to show the new answers at least for the time being. I expect this block post the details in this post to be completely irrelevant in six months time, but at least you will see the differences between how one model and another can differ.
Simplified Model‑Specific Prompting Chart
| Model | Say What You Want vs Don’t Want | How Explicit to Be | Prompt Style That Works Best | Common Failure Mode |
|---|---|---|---|---|
| GPT‑4 / GPT‑4.x | ✅ Strongly prefer what you want | Medium–High | Outcome‑first with clear constraints | Over‑verbosity if length isn’t specified |
| GPT‑4o | ✅ What you want | Medium | Outcome‑first, minimal micromanagement | Adds extra context if prompt is vague |
| GPT‑4o Mini | ✅ What you want (critical) | High (but simple) | Short, focused prompts | Drops constraints when overloaded |
| Claude Sonnet 4.6 | ✅ What you want + alternatives | High | Structured, step‑based prompts | Skips unstated requirements |
| Claude Opus 4.7 | ✅ What you want (very literal) | Very High | Clear success criteria, fewer steps | Won’t infer intent you didn’t state |
| Gemini 1.5 Pro | ✅ What you want | High | Role + task + constraints upfront | Loses focus with mixed intent |
| Gemini 1.5 Flash | ✅ What you want (essential) | Medium | Simple, direct requests | Degrades in long or complex prompts |
| Gemini 2.x | ✅ What you want | Medium | Outcome‑driven with light structure | Still sensitive to ambiguity |
One‑Line Takeaways (for memory)
- All models: Positive instructions beat negative ones.
- Flagship models: Describe success, not steps.
- Smaller models: Fewer rules, clearer focus.
- Claude (especially Opus): If it matters, say it.
- Gemini: Be explicit early; don’t bury intent.
If you want, I can also:
Below are designer‑vibe coding prompt examples, one per model, all for the same task so you can clearly feel the differences.
Shared task
Build a small, stylish UI component (e.g. a settings panel or landing card) with clean design sensibility, sensible defaults, and readable code.
GPT‑4 / GPT‑4.x
Strength: outcome‑first, balanced guidance
Create a responsive settings panel component.
Design goals:
- Calm, modern UI with generous spacing
- Neutral colour palette, accessible contrast
- Clear visual hierarchy
Output:
- React + CSS (or Tailwind)
- Clean, readable code
- Brief explanation of design decisions✅ Why this works
GPT‑4 infers how to get there. You describe what good looks like, not the steps.
GPT‑4o
Strength: designer‑friendly, fast iteration
Design and implement a small settings panel with a strong product-design feel.
Aim for:
- Polished spacing and typography
- Sensible defaults, not flashy
- Feels like something shipped by a mature design team
Deliver:
- React component
- Minimal but clear styling✅ Why this works
GPT‑4o responds well to taste language (“polished”, “mature”) and doesn’t need micromanagement.
GPT‑4o Mini
Strength: simple, explicit, narrow
Create a simple settings panel UI.
- React component
- Light background
- Clear labels and spacing
- No animations
Keep the code short and readable.Requirements:
✅ Why this works
Mini models drop constraints when overloaded—keep it short, concrete, and narrow.
Claude Sonnet 4.6
Strength: structure + clarity
You are a front-end engineer with strong design sensibility.
Task:
1. Design a settings panel UI with calm, modern aesthetics
2. Implement it in React
3. Explain layout and spacing choices briefly
Design constraints:
- Neutral colours
- Clear hierarchy
- Accessible defaults✅ Why this works
Sonnet likes ordered structure and benefits from numbered steps.
Claude Opus 4.7
Strength: literal, precise, high‑fidelity execution
You are designing a production-quality settings panel.
Success criteria:
- Visually calm, modern interface
- Strong spacing rhythm and typography
- Sensible defaults, no unnecessary decoration
Deliverables:
- React component with styling
- Short explanation of design rationale✅ Why this works
Opus will not infer missing intent. Clear success criteria = excellent results.
Gemini 1.5 Pro
Strength: explicit role + task + constraints
You are a product designer who codes.
Task:
- Design and build a settings panel component.
- Audience: Design-conscious product team
Constraints:
- React
- Neutral colour palette
- Clear hierarchy
- Clean, production-ready code✅ Why this works
Gemini Pro performs best when role, task, audience, and constraints are explicit up front.
Gemini 1.5 Flash
Strength: fast, direct, scoped
Create a clean settings panel UI in React.
Focus on:
- Good spacing
- Clear labels
- Simple layout
No explanations. Code only.✅ Why this works
Flash excels at short, direct instructions and degrades with long context.
Gemini 2.x
Strength: outcome‑driven, slightly more flexible
Design and implement a modern settings panel component.
Priorities:
- Calm, tasteful design
- Clear visual hierarchy
- Maintainable React code
Keep it practical and shippable.✅ Why this works
Gemini 2.x handles high‑level outcomes better than earlier versions, but still needs clarity.