Accessing AI sandboxes

1. Click Hands-on in the navigation menu.
2. Click Get Started in the AI Sandboxes tile.

Prompt sandbox

Use the Prompt sandbox to access multiple large language model (LLM) AIs like Jurassic, Claude, and ChatGPT. Practice writing prompts for one or two LLMs and compare the results.

To start a session in Prompt sandbox:

1. Navigate to the AI Sandboxes tab in the Hands-on playground.
2. Click Open Sandbox in the Prompt sandbox card.
3. Click Start Sandbox. This generates a sandbox URL and API key for your use.
4. Select one or two LLMs from the dropdown menus.
5. Type a prompt into the Prompt field. The same prompt will be used for both LLMs, if two are selected.
6. Click the Parameters button next to an LLM to adjust its response, if desired. Parameters vary depending on the LLM(s) selected and include variables such as max response length, temperature, top k, and top p.

   

7. Click Send Prompt.

The response generated by the LLM(s) you selected will appear.

1. You can use the API key generated for your Prompt sandbox session to make HTTP requests such as JavaScript’s fetch. See Prompt sandbox Documentation (opens in new tab) for more information about setting up and using APIs in Prompt sandbox.
2. You’re limited to five Prompt sandbox sessions per day, with each session lasting a maximum of eight hours. After a session expires, you’ll still have access to your previous prompts and responses, but you won’t be able to edit or continue from previous sessions. You can use the Prompt sandbox concurrently with other notebooks or sandboxes.
3. When you start a Prompt sandbox session, you’re provided with a set number of tokens to use towards prompts and responses. You can view the total number of tokens used during your session or per response. Each LLM measures token value differently, and tokens don’t roll over between sessions.
4. The sandbox also provides information on how much your session and each prompt response would cost. This cost isn’t accrued or billed to you or your organization’s plan, but used for demonstration purposes.
5. You can copy the text output of an individual response or use code examples provided by the sandbox.

AI cloud sandboxes

With AI cloud sandboxes, you can spin up a cloud sandbox with access to that cloud provider’s AI services. Like other cloud sandboxes in the Hands-on playground, you’re provided a link to open the cloud console in a private browser window and a set of credentials to log in to the sandbox environment.

To start an AI cloud sandbox:

1. Navigate to the AI Sandboxes tab in the Hands-on playground.
2. Click Open Sandbox in the AI Cloud Sandboxes card.
3. Select a sandbox type from the dropdown menu.
4. Click Start Sandbox. A sandbox URL and credentials will appear.
5. Copy the sandbox URL and paste it into an Incognito or browser-equivalent private window.
   —or—
   Right-click Open Sandbox and open the link in an Incognito or browser-equivalent private window.
6. Copy the user name and paste it into the user name field in the new window.
7. Repeat the above step with the password.

You can only run one cloud sandbox at a time. If you want to switch to a new sandbox or notebook in the Hands-on playground, you need to first end your current session by deleting the AI cloud sandbox you’re using.

AI cloud sandboxes are limited to three hours per session, but there’s no limit to the number of sessions you can open in a single day. Once you delete a session or it expires, you no longer have access to any data or information from the session

Supported services for AI cloud sandboxes

The following are supported services in AI cloud sandboxes and their limits and restrictions.

AWS Bedrock

Limits

• Supported models:
  • AI21 Labs
    Jamba 1.5 Large
    Jamba 1.5 Mini
  • Amazon
    Amazon Nova Canvas
    Amazon Nova Lite
    Amazon Nova Micro
    Amazon Nova Premier
    Amazon Nova Pro
    Amazon Titan Embeddings G1 - Text
    Amazon Titan Image Generator G1
    Amazon Titan Image Generator G1 v2
    Amazon Titan Multimodal Embeddings G1
    Amazon Titan Text Embeddings V2
    Amazon Titan Text G1 - Express
    Amazon Titan Text G1 - Lite
  • Anthropic
    Anthropic Claude v2
    Anthropic Claude 3 Haiku
    Anthropic Claude 3 Opus
    Anthropic Claude 3 Sonnet
    Anthropic Claude 3.5 Haiku
    Anthropic Claude 3.5 Sonnet
    Anthropic Claude 3.5 Sonnet v2
    Anthropic Claude 3.7 Sonnet
    Anthropic Claude Haiku 4.5
    Anthropic Claude Opus 4
    Anthropic Claude Opus 4.1
    Anthropic Claude Sonnet 4
    Anthropic Claude Sonnet 4.5
  • Cohere
    Cohere Command R
    Cohere Command R+
    Cohere Embed English
    Cohere Embed Multilingual
    Cohere Embed v4
  • DeepSeek
    DeepSeek
    DeepSeek-R1
    DeepSeek DeepSeek-V3.1
  • Meta
    Meta Llama 3 8B Instruct
    Meta Llama 3 70B Instruct
    Meta Llama 3.1 8B Instruct
    Meta Llama 3.1 70B Instruct
    Meta Llama 3.1 405B Instruct
    Meta Llama 3.2 1B Instruct
    Meta Llama 3.2 3B Instruct
    Meta Llama 3.2 11B Instruct
    Meta Llama 3.2 90B Instruct
    Meta Llama 3.3 70B Instruct
    Meta Llama 4 Maverick 17B Instruct
    Meta Llama 4 Scout 17B Instruct
  • Mistral AI
    Mistral 7B Instruct
    Mistral Large (24.02)
    Mistral Large (24.07)
    Mistral Small (24.02)
    Mixtral 8x7B Instruct
    Pixtral Large (25.02)
  • OpenAI
    OpenAI gpt-oss-1 20b
    OpenAI gpt-oss-20b
  • Qwen
    Qwen Qwen3 235B A22B 2507
    Qwen Qwen3 32B (dense)
    Qwen Qwen3 Coder 480B A35B Instruct
    Qwen Qwen3-Coder-30B-A3B-Instruct
  • Writer
    Writer Palmyra X4
    Writer Palmyra X5
• Max 50 model invocations
• Max use of 20,000 tokens

Restrictions

• Cannot enable or modify model access agreements
• No access to Bedrock Studio
• Cannot create Bedrock Knowledge Bases

AWS SageMaker

Limits

• Allowed notebook instance types:
  • T-Series, medium to xlarge
  • M-Series, large and xlarge
  • C-Series, large and xlarge
  • R5.large
• Max two SageMaker Notebook instances
• Max two training jobs
• Max one SageMaker endpoint

Restrictions

• Cannot create additional SageMaker Studio environments
• Cannot create clusters, MLFlow Tracking Server, or Fleet
• Cannot run SageMaker Canvas or Data Wrangler
• No accelerators (GPUs) allowed

Azure OpenAI

The normal Azure limits and restrictions also apply to AI cloud sandboxes with the exception of those listed here.

Limits

• Assigned TPM (Tokens Per Minute) capacity of 4,000 (4k)

Restrictions

• The workspace and deployment are automatically created in one of the following supported regions. Learners cannot change, adjust, or manually select the region.
  • Canada East
  • South Central US
  • North Central US
• Cannot create additional deployments
• Cannot access Dall-E image generation
• Cannot access Completions

GCP Vertex AI

Limits

• Supported models:
  • gemini-2.0 Flash
  • gemini-2.0 Flash Lite
  • gemini-2.5 Flash
  • gemini-2.5 Flash Lite
  • gemini-2.5 Pro
  • Text Embedding 004
  • Text Embedding 005
• Max 10,000 (10k) tokens per lab or sandbox
• Existing CPU and memory limits for VMs are applicable to Vertex AI workbench Jupiter notebooks

Restrictions

• No TPUs
• No GPUs

GCP Vertex RAG Corpus

Limits:

• Supported models
  • Text Embedding 004
  • Text Embedding 005

• Max 10,000 (10k) tokens per lab or sandbox

For a full list of supported services in our sandboxes, see our articles on AWS, Azure, and GCP cloud sandboxes.

SageMaker Studio Notebook

SageMaker Studio is an integrated development environment featuring several tools that allow you to explore, clean, and organize data and build, train, and deploy data models. You can also code in multiple programming languages.

Unlike cloud sandboxes, you don’t need to log in or use credentials to access this feature outside of your account credentials. Instead, the platform creates and sets up a pre-created SageMaker Studio Notebook, saving you time and effort.

To start a SageMaker Studio Notebook:

1. Navigate to the AI Sandboxes tab in the Hands-on playground.
2. Click Open Notebook in the SageMaker Studio Notebook card.
3. Click Start Sandbox. A notebook URL will appear.
4. Copy the notebook URL and paste it into an Incognito or browser-equivalent private window.
   —or—
   Right-click Open Notebook and open the link in an Incognito or browser-equivalent private window.