Claude’s API sandbox got a new name and a cost readout

Anthropic has swapped Workbench in the Claude Console for Playground. It’s a stateless sandbox that shows each Messages API call with its tokens, cost and latency, and it stores none of your prompts. “Workbench is now Playground,” says the Console copy quoted in a September 21 Threads post, and the pitch is to “explore the Claude API by using it” and to test models and Messages API features “without writing code.”

You no longer start from an empty box. Playground opens on working examples for tools, citations and caching, and you can load any of its templates and edit them. Code samples in Anthropic’s documentation also have an “Open in Playground” option. Each run shows the request and the response side by side, and once a call does what you want, you copy the SDK code straight out of it.

The help center says Playground is built directly on the public Messages API, so the request you put together there is the same one your code would send. The raw view shows the full message structure, the stop reason and the usage block. That makes the request you debug the same one you ship, with no translation step in between.

Workbench kept a lot more. It saved your work on Anthropic’s servers, where you could version and share prompts, define variables and attach evaluations. Playground keeps only the current draft, and it keeps it in your browser. The help center calls it “a simpler, stateless way to try Claude models and API features in the Console” and says it “does not support saving prompt history or evaluating prompts.” Workbench itself is listed as retired.

This has been underway for a while. A redesign in July moved Workbench into a new Build section, and Anthropic’s release note of July 17 set the cutoff: legacy access would end on August 17, and “saved prompts, variables, and evals are not supported in the updated Workbench.” The new name arrived in August. To get your data out, you used an in-product banner or Organizational Settings, and the export came as a JSON archive of saved prompts, prompt versions, model completions and uploaded files. The export window closed on September 1.

The retirement also hit code that never touched the UI. On August 17 Anthropic shut down three experimental prompt-tool endpoints: /v1/experimental/generate_prompt, /v1/experimental/improve_prompt and /v1/experimental/templatize_prompt. They now return errors, so any pipeline that called them broke. Tech Times reported that Anthropic said there is no way to recover data after the cutoff. It also pointed out that 31 days’ notice is shorter than the minimum of 60 days Anthropic’s documented policy gives before retiring a publicly released model, though that policy doesn’t cover experimental endpoints.

Playground stores nothing on Anthropic’s servers, so the way to keep a prompt you like is the code tab. It exports the current request as a snippet you can put in your own repository.