Service eligibility and regional restrictions
PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.
Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.
Key Takeaways
- On August 18, 2026, Anthropic announced that Workbench in Claude Console was renamed Playground.
- Playground supports all Messages API parameters and provides templates for API features such as code execution and web search.
- Each run displays the full SDK request and API response, helping turn console experiments into reproducible application code.
- The new tool is a lightweight debugging interface without server-side history. It does not save prompts, prompt versions, evals, or shared records. A current browser draft cannot replace version control.
- Access to the old Workbench ended on August 17, 2026. History not exported beforehand cannot be imported through the new Playground. Teams should preserve assets in code repositories, test samples, and evaluation tools.
What Playground is suited for
Playground helps verify Claude models, Messages API parameters, and tool configuration before writing application code. It also shows the full SDK structure and raw response for a request. For developers investigating parameter errors, tool calls, or output formats, it more closely reflects real API behavior than a chat interface alone.
It is not a team prompt repository, long-term conversation archive, or formal eval platform. Teams needing collaboration, version rollback, batch evaluation, and auditing should put exported code, inputs, expected results, and metrics into their own repositories and testing processes.
Key differences from the old Workbench
The old Workbench could save prompts, prompt versions, completions, variables, evals, and shared records. Playground uses a simpler, stateless experience and does not save this work on Anthropic's servers. The current draft may remain in the browser, but clearing browser data, changing devices, or switching accounts may lose it.
The new interface is built directly on the public Messages API and displays complete requests and responses. The name change is not a model upgrade and does not automatically change API keys, workspaces, model permissions, quotas, or pricing.
Can old data still be migrated?
Access to the old Workbench ended on August 17, 2026. Anthropic previously allowed users to download prompts, revisions, completions, uploaded files, and evals through Export data in the old interface. Those saved contents are no longer accessible after retirement.
Exported data cannot be imported into Playground because the new tool has no server-side history storage. If your team already has an export, preserve the original JSON first, then organize core prompts, variables, model parameters, input samples, and expected outputs in version control. Do not assume official support can restore history that was never exported.
From an experiment to a reproducible request
First select a model and enter system and user content, then configure max tokens, temperature, thinking, tools, or other Messages API parameters for the task. After running, inspect the visible output, full API response, stop reason, usage, and tool calls together. Do not copy only the final text.
Use the SDK request or code export as a starting point and place the API key in an environment variable rather than source code. Rerun locally with the same model, parameters, and input. Record the model ID, SDK version, request ID, input snapshot, and expected assertions to determine where console and application results diverge.
Using code execution and web search templates
Playground provides templates for API features such as code execution and web search to demonstrate parameters and request structure. Templates are learning aids, not production security configurations. Before enabling a tool, verify model support, beta headers, permissions, cost, timeouts, data boundaries, and failure handling.
After bringing a template into your application, test at least unavailable tools, timeouts, empty results, refusals, parameter-validation failures, and repeated calls. Retain sources for external search results, constrain execution-tool inputs and outputs, and keep API keys and sensitive data out of prompts and logs.
Four common misconceptions
The first is treating Playground as a chat product and overlooking its connection to the Messages API. The second is assuming a rename automatically preserves old history. The third is treating a current browser draft as a reliable backup. The fourth is assuming one console success guarantees the production SDK, network, permissions, and concurrency will also work.
If your application returns 401, 404, 429, or 5xx, first investigate status code, model name, permissions, quota, and retry strategy. See the AI API configuration guide and AI API error troubleshooting guide. Move to network troubleshooting only when there is evidence of DNS, TLS, proxy authentication, or connection timeouts.
Team migration and acceptance checklist
Check whether old data has an exported copy, core prompts are version-controlled, variables and keys are separated, model and parameters are fixed, real samples have expected results, SDK requests can be rerun locally, tool failures are handled, and cost and latency have baselines. If anything is missing, do not connect a demonstration request directly to production.
Teams needing stable access to Claude Console, API documentation, and overseas development tools can visit the PuppyIP website to learn about fixed network exits. Network conditions affect only the connection path. They cannot restore old Workbench data or change model permissions, quotas, or product availability.
Sources
Frequently Asked Questions
Are Claude Workbench and Playground two different products?
On August 18, 2026, Anthropic renamed Workbench in Claude Console to Playground. This updates a console tool's name and experience; it is not a new model.
Does Playground save prompt history and evals?
No. The new tool does not save prompts, prompt versions, evals, or shared records on the server. A current draft remaining in the browser is not a long-term backup.
Can old Workbench data be imported into Playground?
No. Data could be exported before the old tool retired, but Playground has no history storage to import into. Move existing exports to your own repository or evaluation system.
Which API parameters does Playground support?
Official guidance says it supports all Messages API parameters and provides feature templates such as code execution and web search. Check documentation for each model's compatibility and beta requirements.
Why does Playground work while my application code fails?
Common differences include the API key, workspace, model ID, SDK version, parameters, beta headers, quota, and network. Compare the full SDK request and API response layer by layer.
Can Playground output directly serve as production acceptance?
That is not recommended. One success proves only that the current request runs. Before production, use real samples to check quality, cost, latency, tool failures, concurrency, and rollback.