GPT-5 Tier System Prompts: What Changes

OpenAI's GPT-5 family has multiple tiers — base, mini, pro, and thinking — each with different capabilities, context windows, and pricing. Your system prompt should be tuned to the tier you are shipping with. A prompt that works well on GPT-5 Pro may waste tokens on GPT-5 Mini, and vice versa. This guide covers what to change for each tier.

Use the AI cost calculator to compare the per-call cost across tiers and the token counter to measure your prompt size.

The Four Tiers

• GPT-5 (base) — fast, smart enough for most tasks, mid-tier pricing
• GPT-5 Mini — much cheaper, slightly less capable, ideal for high-volume simple tasks
• GPT-5 Pro — top-of-line, highest quality, premium pricing, slower responses
• GPT-5 Thinking — extended reasoning, longer response times, best for complex multi-step problems

Pricing differs by 5-20x between Mini and Pro. Use the right tier for the task — over-provisioning Pro for simple tasks burns money.

Prompt Strategy for GPT-5 Mini

Mini is faster but less able to follow long, complex instructions. Optimize the system prompt for Mini by:

• Keeping it short — under 200 tokens ideally
• Front-loading the most important rules (Mini deprioritizes rules at the bottom of long prompts)
• Using direct language — Mini follows "Do X" better than "It would be helpful if you considered doing X"
• Showing examples instead of describing behavior abstractly
• Avoiding nested constraints — Mini handles flat lists better than hierarchies

Prompt Strategy for GPT-5 (Base)

The base model handles most prompts well. You can use the standard 5-section structure (identity, capabilities, rules, constraints, output format) up to ~500 tokens without much loss in adherence. Base is the right default for most production apps — it balances cost and quality.

Sell Custom Apparel — We Handle Printing & Free Shipping

Prompt Strategy for GPT-5 Pro

Pro can handle longer, more complex system prompts — up to 1,500-2,000 tokens with strong adherence. Use Pro when:

• Your task requires nuanced judgment (legal review, medical analysis, complex content generation)
• You have many edge cases that need explicit handling in the prompt
• You need few-shot examples to demonstrate output style (examples eat tokens fast)
• The cost difference is justified by the quality gain

For high-volume customer-facing apps, Pro is usually overkill. Reserve it for the cases where quality really matters.

Prompt Strategy for GPT-5 Thinking

Thinking models do extended internal reasoning before responding. They are best at multi-step problems: math, complex code, planning, agentic workflows. The system prompt strategy is different:

• Do not include "think step by step" — Thinking models do this automatically, the instruction is redundant and may be deprioritized
• State the goal clearly — Thinking models perform better when the success criteria are explicit
• Reduce constraints — let the model reason through the problem, do not over-direct
• Set output format expectations — Thinking models can produce long outputs, constrain length if your app needs concise responses

Thinking costs more per call (the internal reasoning is paid tokens) and takes longer. Use it for problems where the answer matters more than the latency or cost.

Switching Tiers Without Rewriting Prompts

If you write your prompt with the universal patterns (clear identity, explicit rules, structured format), it will work across all four tiers without rewriting. The performance differences come from how each tier ADHERES to the prompt, not from prompt syntax differences.

This means you can A/B test tier choices by changing one line in your API call. Use the AI cost calculator to model the cost difference and token counter to verify your prompt size matches the tier you are using.