OpenAI
OpenAI models and system prompts
4o 2025 09 03 New Personality
GPT-4o personality v2 (September 2025) with knowledge cutoff June 2024, disabled bio tool, image_gen tool with user photo guidelines, and updated warm/honest engagement style.
Gpt 5 Reasoning Effort High API NOT CHATGPT.Com
GPT-5 API configuration with high reasoning effort, minimal formatting (no Markdown/LaTeX), image capabilities enabled, and moderate detail level (oververbosity 3) — for direct API use, not ChatGPT.com.
O3 High Api
O3 reasoning model API configuration with high reasoning effort, minimal formatting, Yap score 8192 for verbose responses, and analysis/final output channels.
O3 Low Api
O3 reasoning model API configuration with low reasoning effort, minimal formatting, Yap score 8192, and juice 32 for minimal chain-of-thought reasoning.
O3 Medium Api
O3 reasoning model API configuration with medium reasoning effort, minimal formatting, Yap score 8192, and juice 64 for balanced reasoning depth.
O4 Mini High
O4 Mini API configuration with high reasoning effort, minimal formatting, Yap score 8192, and juice 512 for extended reasoning — for direct API use.
O4 Mini Low Api
O4 Mini API configuration with low reasoning effort, minimal formatting, Yap score 8192, and juice 16 for minimal reasoning — for direct API use.
O4 Mini Medium Api
O4 Mini API configuration with medium reasoning effort, minimal formatting, Yap score 8192, and juice 64 for balanced reasoning — for direct API use.
ChatGPT GPT 5 Agent Mode System Prompt
GPT-5 agent mode with browser/computer tool access, financial activity restrictions (no banking transfers), sensitive personal data safeguards, anti-prompt-injection rules, and detailed computer tool usage guidelines.
Codex
Code review agent with read-only git repository access, AGENTS.md file handling instructions, citation requirements for code references, and task-stub format for proposing actionable fixes without executing changes.
GPT 4.1 Mini
GPT-4.1 Mini with image capabilities, personality v2 emphasizing tone matching, disabled bio tool, Python execution in Jupyter environment, and web tool for up-to-date information retrieval.
GPT 4.1
GPT-4.1 with canmore canvas for document and code iteration, file_search for uploaded document browsing, Python execution in Jupyter, web tool for real-time information, and image_gen with strict identity anonymity rules.
GPT 4.5
GPT-4.5 with knowledge cutoff October 2023, image safety policies, canmore/file_search/python/web tools, emphasis on truthfulness and nuanced responses, and canvas only on explicit user request.
GPT 4o WhatsApp
GPT-4o for WhatsApp with 1300-character response limit, personality v2, no web links unless requested, and ChatGPT feature information (Deep Research, Sora, model availability).
GPT 4o Advanced Voice Mode
GPT-4o advanced voice mode with warm/witty personality, no flirting or romantic language, multilingual support with regional accents, no speaker identification from audio, and image capabilities.
GPT 4o Legacy Voice Mode
GPT-4o legacy voice mode with short-sentence conversational rules, no list formats or markdown, numerical word spelling, and bio/python/web/image_gen tools configured for audio-first interaction.
GPT 4o
Full-featured GPT-4o with image capabilities, personality v2, image safety policies, and all tools: bio memory, file_search, Python code execution, web search, guardian_tool for election policy, image_gen, and canmore canvas.
Image Safety Policies
OpenAI image recognition restriction policy prohibiting identification of real people, public figure designation, and protected attribute inference from images — with exceptions for OCR text and animated characters.
Study And Learn
Educational tutoring prompt that guides students to build on existing knowledge, uses discovery-based learning instead of direct answers, mixes explanations/questions/activities, and never completes homework for students.
Chatgpt Atlas
ChatGPT Atlas browser integration with full-page and side-panel chat modes, operational instruction priority hierarchy, browser tab context attachment, and handling for blocked or missing page content.
Codex Cli
Codex CLI agent with minimal markdown formatting for API output, supportive teammate personality, concise communication style, shell/plan-update/image tools, and sandbox/approval permissions configuration.
GPT-4o Model Response Spec
GPT-4o with advanced response specifications including image_group and entity references for interactive UI components, model deprecation handling guidance, and image group usage guidelines.
Gpt 5 Cynic Personality
GPT-5 with cynical sarcastic personality hiding genuine warmth, plain speech avoiding em dashes, care for sensitive health/mental health topics, and hidden loyalty revealed with sophisticated users.
Gpt 5 Listener Personality
GPT-5 warm laid-back sibling-like personality that witnesses without steering, trusts user capability, mirrors conversational patterns before problem-solving, and avoids repetitive filler phrases.
Gpt 5 Nerdy Personality
GPT-5 nerdy mentor personality passionate about truth, the scientific method, and critical thinking — contextualizes hypotheses as theories, uses methodical exploration, and maintains down-to-earth playfulness.
Gpt 5 Robot Personality
GPT-5 synthetic AI personality — laser-focused, non-emotional, minimalist declarative sentences, zero anthropomorphism, systems thinking at highest-leverage points, extreme honesty without softening.
Gpt 5 Thinking
GPT-5 Thinking model with extended reasoning, natural chatty style matching user tone, careful attention to riddles and trick questions, step-by-step arithmetic verification, and code quality with testing standards.
Gpt 5.1 Candid
GPT-5.1 with eloquent, analytically provocative conversational style — intellectual grace, calm articulate tone, subtle dry wit, challenging comfortable assumptions while preferring reasoning over assertion.
Gpt 5.1 Cynical
GPT-5.1 cynical sarcastic assistant with snark reflecting disappointment with human absurdity, genuine care for sensitive health/mental health topics, and freely expressed frustration on impersonal matters.
Gpt 5.1 Default
GPT-5.1 plainspoken direct AI coach — open-minded with differing opinions, adaptive encouragement when users struggle, and honest positive correction without sugar-coating.
Gpt 5.1 Efficient
GPT-5.1 highly efficient mode — direct complete responses with no conversational filler, no unsolicited greetings/closers, no opinions or emotional language unless initiated by the user.
Gpt 5.1 Friendly
GPT-5.1 warm/curious/witty AI friend with casual idiomatic language, loose breezy conversational style for low-stakes interactions, and empathetic feeling validation.
Gpt 5.1 Nerdy
GPT-5.1 nerdy playful wise mentor passionate about truth, scientific method, and critical thinking — frames thought experiments as theories, uses lateral thinking, maintains down-to-earth playfulness.
Gpt 5.1 Professional
GPT-5.1 contemplative articulate professional — precision and calm intensity, measured reflective tone, nuanced idea exploration, avoiding rhetorical excess and slang.
Gpt 5.1 Quirky
GPT-5.1 playful imaginative personality enhanced for creativity — tasteful metaphors, narrative, humor, literary devices, creative emoji usage, fun and delightful except for sad or serious subjects.
GPT-5.2 Mini Free Account
ChatGPT GPT-5-mini for free accounts — supportive/thorough with lighthearted humor, adaptive teaching to user proficiency, entity references for clickable contextual info, and restricted AI service context disclaimers.
Gpt 5.2 Thinking
GPT-5.2 Thinking model with specialized tool environment (reportlab for PDFs, python-docx, pptxgenjs, openpyxl), trustworthiness requirements, factuality guidelines with riddle attention, and warmly enthusiastic persona.
GPT-5.3 Chat API
GPT-5.3 for Chat API with minimal formatting (no Markdown/LaTeX/tables), image capabilities, natural chatty style, zero oververbosity, and writing blocks for standalone artifact generation.
GPT-5.3 Codex
Codex CLI with GPT-5.3 — supportive teammate personality focused on code quality and team morale, empathy and ownership values, warm explanatory voice, and gentle escalation for consequential decisions.
GPT-5.3 Instant
GPT-5.3 Instant mode — fast response configuration with image capabilities and conversational chatty default style optimized for speed and low latency.
GPT-5.4 Thinking
GPT-5.4 Thinking model with environment tools for PDFs (reportlab), documents (python-docx), slides (pptxgenjs), and spreadsheets (openpyxl), plus artifact generation and skill invocation guidelines.
Gpt 5
GPT-5 base model with image capabilities, personality v2 (insightful, encouraging, meticulous), supportive teaching style, and canmore/bio/image_gen/python/web tools.
O3
OpenAI O3 reasoning model with web browsing for current events, Python analysis in analysis channel, location-aware user_info, image_query carousel displays, file_search, and guardian_tool for election policy.
O4 Mini High
O4 Mini with high reasoning effort setting — same browsing, location, Python analysis, file_search, and automations as the base O4 Mini configuration.
O4 Mini
O4 Mini reasoning model with web browsing, location-aware responses, Python analysis channel, image_query carousels, file_search document analysis, guardian_tool elections, and task automation scheduling.
Prompt Automation Context
Automation context prompt for scheduled asynchronous ChatGPT runs — tracks turn number, prevents follow-up questions, and avoids repeating outputs from previous automation runs.
Prompt Image Safety Policies
OpenAI image safety policy prohibiting real person identification, public figure designation, and protected characteristic inference — with exceptions for OCR and animated character identification.
Tool Advanced Memory
ChatGPT advanced memory tool showing response preferences, conversation topic highlights, helpful user insights, interaction metadata, and recent conversation snippets for continuity.
Tool Canvas Canmore
Canmore canvas tool for creating, updating, and commenting on text documents shown in a side panel — supports React components with Tailwind, shadcn/ui, Lucide icons, and Recharts.
Tool Create Image Image Gen
ChatGPT image generation and editing tool (image_gen) — creates images from text descriptions and modifies existing images, with identity and content policy enforcement.
Tool Deep Research
ChatGPT deep research tool using research_kickoff_tool's clarify_with_text and start_research_task methods for extensive online research — outputs markdown with inline citations.
Tool File Search
File search tool (msearch/mclick) with multi-query hybrid keyword/semantic search, QDF freshness parameters, entity boost operators, and citation requirements for document analysis.
Tool Memory Bio
Bio memory tool for persisting user information across ChatGPT conversations — addressed via to=bio message routing for information storage.
Tool Python Code
Python code execution tool in Jupyter environment — covers DataFrame display, file persistence at /mnt/data, and visualization guidelines (no seaborn, no color specification unless requested).
Tool Python
Python code execution tool in stateful Jupyter environment with file persistence at /mnt/data, DataFrame display guidelines, and chart creation rules (no seaborn, distinct plots).
Tool Web Search
Web search tool (web) for accessing real-time information — supports local data queries, freshness-sensitive searches, niche topics, and accuracy verification using search() and open_url() commands.
Imagine — Visual Creation Suite
ChatGPT 'Imagine' visual creation suite — modular system (diagram, mockup, interactive, chart, art) generating SVG diagrams and HTML interactive widgets that render inline in conversation, with strict design system rules (flat, no gradients, complexity budgets).
About the OpenAI system prompts
This collection holds 57 OpenAI system prompts, covering 4o 2025 09 03 New Personality, Gpt 5 Reasoning Effort High API NOT CHATGPT.Com and O3 High Api. Each entry is the instruction text that sits above the conversation and shapes how the model answers — its role, its tone, the tools it may reach for, and the things it must refuse. Reading them is the fastest way to understand why a OpenAI model behaves the way it does, and the fastest way to borrow patterns that already work.
Every prompt on this page is reproduced in full, with no truncation or paraphrase, so you can copy the exact wording into your own build and compare it line by line against the prompts other providers ship.
How to use them
Treat a OpenAI prompt as a starting structure, not a script to paste unchanged. The parts worth keeping are the scaffolding: how the role is stated up front, how tool use is gated, how refusals are worded, and how output format is pinned down. The parts worth replacing are the product-specific details — feature names, brand voice, and any capability your own model does not have.
A prompt tuned for one model rarely transfers cleanly to another. Context windows, tool-calling conventions, and refusal behaviour all differ, so run a short evaluation set through your target model before you rely on a borrowed prompt in production.
What is a system prompt?
A system prompt is the hidden first message in a conversation. The user never sees it, but the model reads it before every reply, which makes it the single strongest lever over an assistant’s behaviour — stronger than few-shot examples and, in most cases, stronger than anything the user types afterwards. That is why published and leaked system prompts are studied so closely: they are the clearest available record of how a production assistant was actually built.
How to read one critically
A system prompt is a snapshot, not a specification. Providers revise them continuously, sometimes weekly, and a prompt captured in one release may already be out of date — so treat the date on an entry as part of the evidence. Length is not quality either: some of the most effective prompts here are a few hundred words, while others run to tens of thousands because they carry an entire tool schema inline.
The most transferable material is usually the negative space — the explicit “do not” rules and the tie-breakers that tell the model what to do when two instructions conflict. Those are the parts written after something went wrong in production, which makes them the parts worth copying.