AI Playground Quick Start Guide

How to enter a prompt: 

1. To use the default settings with your prompts, navigate into the empty field at the bottom of the welcome screen.
2. Type a prompt into the field and press the Return or Enter key.
3. Once you receive the result, continue to modify or refine your prompt.
   • See guidance.

Attach files

1. Next to the box where you enter your prompts, appears a paperclip icon.
2. Press the icon to open a window in which you can select the file you wish to upload.
3. Type out your prompt as the file uploads.
4. When the file is finished uploading, and you have finished writing your prompt, simply press the return or Enter key.

Please note, the paperclip icon does not appear when using agents.

Interacting with a message in the conversation

Beneath your prompts and the responses generated by the LLMs, you'll find several more advanced features: 

1. Read aloud: Read outs the message via a computer-generated voice.
2. Edit: Allows you to edit your prompts as well as the responses of the various models.
   • Save & Submit: Saves your edit and resubmits the information to regerate the AI's response.
   • Save: Saves your edit without regenerating the response.
   • Cancel: Closes the edit window without saving changes.
3. Copy to clipboard: Copies the content of the selected message to your clipboard to be pasted into another window or program.
4. Regenerate: Forces the model to try to create a new response without any additional context.
5. Fork: Creates a new conversation that starts from the specific message selected. This can be useful for refocusing the conversation, creating branching separate scenarios, preserving context, and more. 

Choose between these available versions for each model.

Azure OpenAI (ChatGPT) models

Model	Best for	Strengths	Notes	Use if you need
gpt-5	Complex reasoning, document processing (text, images, PDFs), long form synthesis, and advanced coding	Strongest reasoning and instruction following capabilities, reliable with longer context, high quality writing, and code generation	Higher cost and much slower than other GPT models, still benefits from structured prompts and examples	Intricate work for technical, policy, research, data analysis, coding, mathematics, and executive summaries
gpt-5-mini	Quick responses and drafts, chat assistants, lightweight code, and email summarization	Quick with and good reasoning, better grounding than previous mini models	Trades depth for speed, not ideal for long discussions or large documents	Quick answers, note cleanup, triage, batch content generation, chatbots, simple agents, rapid prototyping under tight budgets
gpt-4o-mini	Speedy responses, lightweight content generation, casual coding support	Cost-effective, solid for general tasks	Think of it as a faster, trimmed-down gpt-4o with solid performance and less nuance	Speed and value over cutting edge intelligence
gpt-4o	Complex reasoning, image and PDF analysis, advanced code generation	Deep comprehension, contextual awareness, handles vision input natively	Fast and high-performance across creative and technical domains	Advanced problem-solving, file analysis, data visualization, coding, long essays
gpt-4.1	Complex reasoning, professional communications, data analysis, and code generation	Multimodal capabilities, natural conversation flow, coding and technical support	May occasionally generate incorrect or outdated information, needs clear prompts for image generation	A smart creative or professional collaborator, help explaining complex topics simply, fast and reliable assistance on code or data tasks
o1	Academic use, symbolic math, structured data analysis	Especially good at formal reasoning and tough equations	Slower than gpt-4o, but built for precision in logical tasks	Math derivations, proofs, or rigorous academic writing
o3-mini	Quick responses to technical questions	Optimized for STEM, excels at coding, math equations, and science questions	Great utility model for short form technical queries	Quick help with formulas, equations, code snippets

 

Google models

Model	Best for	Strengths	Notes	Use if you need
gemini-2.0-flash-lite-001	Very quick responses, great at back and forths conversations	Speed focused, ideal for chatbots and mobile use	Sacrifices complexity and subtlety for responsiveness	Fast responses and do not require deep reasoning
gemini-2.0-flash-001	Quick language translation, friendly emails, and strong technical reasoning	Handles idioms, coding, high context tasks	Sweet spot for blend of speed + sophistication at lower price	Multilingual understanding and real-time support
gemini-2.5-pro	Complex reasoning, advanced coding, multimodal analysis	State-of-the-art performance, massive context window (1 million tokens), adaptive reasoning, rich multimodal input support	Heavy compute and latency due to deep thinking, still in beta, susceptible to hallucinations	Deep research and document analysis, complex web development, reviewing videos, reading charts

Anthropic models

Model	Best for	Strengths	Notes	Use if you need
claude-3-7-sonnet	Code analysis, deep reasoning, and summarizing long text	Larger context windows allows model to see more data at once	Solid general-purpose model with strong alignment across categories	Help with legal documents, research papers, and deep Q&A on longer texts
claude-3-5-sonnet	Creating large bodies of creative or academic writing	Nuanced, writer friendly model with higher context limit	Excellent for research papers, detailed reports, or complex narratives	Essays, storytelling, or reviewing large documents
claude-3-haiku	Fast processing of structured tasks and existing data sets	Efficient and great at for form fill tasks with clean outputs	Use it for quick instruction following or form based responses	Snappy summaries, quick data extraction, JSON handling
claude-4-sonnet	Everyday coding tasks, writing and document review, reading text and images	Efficient and fast, large context support, strong reasoning	Can be compared to Gemini-2.5-pro, still prone to hallucinations	Coding assistance, analyzing screenshots, documents, and writing
claude-4.5-sonet	Deep reasoning, long running tasks, and advanced coding workflows	Handling complex, multistep projects with consistency	Most accurate and detailed model from Anthropic	Software development, data analysis, research, and agent orchestration

DeepSeek models

Model	Best for	Strengths	Notes	Use if you need
deepseek-r1	Coding, technical Q&A, advanced math reasoning	Open model which gives you insight into its "thinking" process	If you want an open alternative to gpt-4, this is a good pick with verbose results	Help with engineering tasks, logic puzzles, or lots of information on a single topic
DeepSeek-V3-0324	Code generation, complex reasoning, academic-style Q&A	Quality responses across logic heavy and technical domains, excels in benchmark style tasks	Stronger performance in math and step by step reasoning than earlier DeepSeek models	Reliable model for technical asks, code heavy workflows, or educational content generation

Meta models

Model	Best for	Strengths	Notes	Use if you need
Llama-3.2	Adapting writing style and as well as multilingual tasks and translation	Open source and responds well to various types of input	Capable of great instructional clarity with fast responses	Assistance with writing tasks and multilingual support
Llama-4	Open source alternative to more expensive models, especially for writing and multilingual tasks	Strong performance in summarization, translation, and context sensitive outputs	More capable than Llama 3, with improvements in consistency and language understanding	Open source model for writing, creative tasks, and research support across languages

To compare the result of two models simultaneously for the same prompt, use the model compare feature. 

How to compare models side-by-side:

1. Select one model for your comparison in the top menu drop-down options.
2. Click the plus icon in the top menu options.
   • You will notice the selected model is indicated in the prompt field.
3. Select the second model for your comparison in the top menu.
4. Type and enter your prompt into the prompt field.
5. You will see both results side-by-side.
6. Once you click out of the conversation, you will be able to see both responses by tabbing back and forth using the numbered tool below the conversation.

Azure Assistants

Azure Assistants function similarly to traditional chatbots but offer enhanced capabilities. Powered by Azure OpenAI, these assistants do not yet support the full range of advanced features available through integration with external tools. 

At this time, the only available Azure Assistant is:

Data/Code Analyst

• Best for: Processing files with diverse data and formatting likes spreadsheets and PDFs
• Strengths: Excels at solving challenging code problems and provides more robust data analytics capabilities
• Notes: Power by OpenAI’s code interpreter to help improve data analysis over conventional models
• Use if you need: Assistance reviewing complex spreadsheets and PDFs with unstructured data

Agents

Previously known as plugins, agents are similar to Azure Assistants in their ability to perform tasks like coding, searching documents, or generating images. However, agents offer greater customization and can integrate a larger variety of third party capabilities and model providers.  

Available agents include:

• Google Gemini 2.5 Flash Image (aka nano-banana): This image generation tool produces images from text prompts. Especially useful for rapid prototyping, brainstorming, or when you need fast image generation without sacrificing overall quality. (Not available to students, postdocs, or visiting scholars at this time.)
• Imagen 3: Generates exceptional photorealistic images based on text descriptions.(Not available to students, postdocs, or visiting scholars at this time.)
• Dall-E 3: Turns your natural text prompts into AI generated images.(Not available to students, postdocs, or visiting scholars at this time.)
• Google: An AI-assisted Google web search. You can use it paired with the GPT models to help conduct Google searches through the lens of an AI agent.
• ScholarAI: Helps you find and analyze academic papers with ease. This agent can help you streamline research efforts and gain insights faster. Learn more about the feature set and prompting suggestions for this agent.
• DoResearch Policy guide: Allows for AI assisted searching of Stanford's comprehensive collection of policies, guidelines, and general information related to the university's research activities.
• Admin Guide Search: Can be used to help answer simple questions about the Stanford Admin Guide through the lens of an AI model.
• Wolfram: Provides computational intelligence for solving complex mathematical equations.
• Faculty Handbook Search: Designed to help you find specific information related to Stanford University faculty policies and guidelines.
• AI Helper agent: For quick answers about the AI Playground, how to get more out of AI, and for help identifying areas where AI can help support your day to day tasks. 
   

Adjust model parameters

You can also use the configuration options button to the right of the selected model to customize your settings.

Before entering your prompt, you can choose to modify the selected model's settings. This is optional, and most people can leverage the default options for the best experience. 

Note that the settings might vary slightly for each model type, but many models will have the below configuration options: 

• Max context tokens - Defines maximum input size. (1000 tokens is about 750 words)
• Max output tokens - Defines maximum output size. (1000 tokens is about 750 words)
• Temperature - Controls the “creativity" or randomness of the content being generated.
• Top P - Alternative to temperature, refines the number of possible responses based on context provided.
• Reasoning effort - For some models, increasing reasoning effort can help increase quality of responses, by indicating that more computational effort is required. Increased reasoning effort might increase the amount of processing time, as well.

Use bookmarks to organize conversations

Use the Bookmark tool to organize your conversations by similar topics, to find them easily in the future.

Step-by-step:

1. When viewing a conversation, click the bookmark icon in the top bar.
2. Click New Bookmark.
3. Fill in the Bookmark details and click Save.
   • To apply the bookmark to the current conversation (likely desired), check the box to “Add to current conversation.”
4. Now you can use the Bookmarks selector in the left-side panel to view only conversations with the applied bookmarks.
5. Manage, rename, and delete bookmarks in the right-side panel under the Bookmarks section.

You have the ability to share conversations you have had with the models. Your name and any messages you add to the conversation after creating the link stay private.

Share a link to a conversation:

1. Click Share.
2. Click the Create link button to generate a shareable link.
   • Note: Be careful when using this feature. While you must log in to view the link, the conversation will become accessible to any authenticated Stanford users with the link.

Remove a link to a shared conversation:

1. Click on your own name in the bottom left corner.
2. Click Settings.
3. In the new pop up window, click Data controls.
4. Next to Shared links, click on the Manage button.
5. This will display all the chats with shared links, the date shared, and give you an option to delete the link.
   • Note: Deleting a shared link is a permanent action and cannot be undone. Resharing the conversation would include any new information input or generated in the conversation since the original link was generated. 

Explore settings to customize options that impact your entire AI Playground experience.

To access the settings menu:

1. Click on your own name in the bottom left corner.
2. Click Settings. 

General settings

• Theme: Allows you to change between Light and Dark mode.
• Auto: Scroll to latest message on chat open - When enabled, this will automatically move your view to the last message in the conversation.
• Hide right-most side panel: When enabled, this will remove the pop up side panel menu.
• Archived chats: Allow you to unarchive conversations or delete them from the system entirely. 

Chat settings

• Message Font Size: Adjust the size of fonts used in the AI Playground.
• Chat direction: Change the direction of writing systems used in the AI Playground.
• Press Enter to send message: When enabled, pressing the Enter key will send your message.
• Maximize chat space: Increases the visible space used by conversations.
• Open Thinking Dropdowns by Default:
• Parsing LaTeX in messages: When enabled, thinking dropdown menus are expanded automatically. Currently only works with DeepSeek models.
• Save drafts locally: When enabled, texts and attachments you enter in the chat will be saved locally as drafts. Drafts are deleted once the message is sent.
• Default fork option: Defines what information is visible when forking conversations.

Beta features

• Toggle Artifacts UI: Show or hide the Artifacts panel to view, manage, or interact with generated outputs like files, code, and designs.
• Include shadcn/ui instructions: Enable guidance and usage tips or examples for integrating or customizing shadcn/ui components directly in your workflow.
• Custom Prompt Mode: Enable to craft and run highly tailored prompts with full control over structure, behavior, and formatting—ideal for advanced or repeatable tasks.

Commands

Turn on these command shortcuts to help streamline prompt configuration and boost productivity.

• Toggle command “@” for switching endpoints, models, presets, etc.
• Toggle command “+” for adding a multi-response setting
• Toggle command “/” for selecting a prompt via keyboard

Data controls

• Import conversations from a JSON file: Allows you to import conversations exported from other GPT chat applications.
• Shared links: Allows you to view and delete all shared conversations under your account.
• Clear all chats: Deletes all conversations from the left most side panel. (Does not delete archived conversations.) 

Account settings

• Profile picture: Allows you to upload a profile picture for yourself, which is shown in your conversations with the AI models.  (Image must be under 2MB.)
• Display username in messages: When enabled, your name is shown next to your prompts in your conversations.  When disabled, prompts you send will be labeled as "You" in conversations. 

Explore the ability to generate code and corresponding prototypes using React, HTML5, three.js, WebGL, and more.

How to enable Artifacts UI:

1. Click your name in the lower left corner.
2. Open Settings.
3. Click Beta features.
4. Click to switch on the Toggle Artifacts UI.

Note: Artifacts work with each of the major models, but tends to work best with Anthropic and Azure OpenAI models. The artifacts feature does not work in conjunction with agents at this time.

The AI Playground is continually evolving. Features are being updated and added regularly. (View AI Playground release notes.) Here you will find details for using several recently released features.

Accessibility Improvements

• Added support for model specification descriptions.
• Added label accessible names to buttons for file attachment removal.
• Enhanced accessibility and visual consistency, updating base styling and labels in multiple places.
• Refactored the SearchBar and conversation list to remove redundant hooks for better maintainability.
• Enhanced keyboard and screen reader accessibility of the SearchBar and clear-search button, including appropriate labels, focused state, and tabbable logic.
• Improved "Fork" and "SplitText" accessibility design.

AI API Gateway

This new service offers developers, researchers, and tinkerers direct API access to the models within the AI Playground. The service requires a valid PTA to use. You can learn more about this new service on the service page and request an API key to get started. 

New AI Agents

• Admin Guide Search: Can be used to help answer simple questions about the Stanford Admin Guide through the lens of an AI model.
• DoRearch Policy Guide: Allows for AI-assisted searching of Stanford's comprehensive collection of policies, guidelines, and general information related to the university's research activities.
• Faculty Handbook Search: Designed to help you find specific information related to Stanford University faculty policies and guidelines.

Open Thinking Dropdowns (DeepSeek models only)

Using the "Open Thinking" feature with DeepSeek models, you’ll be able to see real-time which criteria and inputs the model is considering as part of its response. This allows you to improve your prompts and better tailor follow-ups. You can make sure this content is automatically expanded in Settings.

To enable, go to Settings > Chat > Open Thinking Dropdown by Default. The option currently only works for the DeepSeek models at this time.

Temporary Chat

A new feature called Temporary Chat can be used:

• To keep your chat history clean and focused
• For sensitive topics, quick experiments, or anything you don’t need to permanently save.

These chats are excluded from your personal search results, cannot be bookmarked, and are automatically deleted after 30 days.