Previous Card
LEANN: A Low-Storage Vector Index
GPT-5 prompting guide
This guide provides prompting tips to maximize the quality of GPT-5's outputs, focusing on improving agentic task performance, ensuring instruction adherence, utilizing new API features, and optimizing coding for various engineering tasks. It emphasizes calibrating the model's eagerness, using tool preambles, and leveraging the Responses API for efficiency. ✨
Article Points:
1
GPT-5 excels in agentic tasks, coding, raw intelligence, and steerability.
2
Prompting tips maximize GPT-5 output quality for agentic tasks and coding.
3
Calibrate GPT-5 eagerness via `reasoning_effort` and explicit prompts.
4
Tool preambles provide clear progress updates in agentic workflows.
5
Responses API improves agentic flows, lowers costs, and reuses reasoning.
6
Optimize coding by adhering to standards and using self-reflection.
Source:
GPT-5 prompting guide
Core Capabilities
Agentic Task Performance
Coding Excellence
Raw Intelligence
Steerability
Prompting Strategies
Agentic Eagerness
- Control with reasoning_effort
- Define context gathering
- Encourage persistence
Tool Preambles
- Provide clear upfront plans
- Offer consistent progress updates
Coding Optimization
- Frontend app development
- Match codebase standards
- Self-reflection rubrics
Instruction Following
- Control verbosity parameter
- Resolve conflicting instructions
API Features
Responses API
- Reuses reasoning context
- Improves agentic flows
- Lowers costs
Advanced Techniques
Prompt Optimizer Tool
Metaprompting
Source: