Anthropic’s Playground vs. OpenAI’s: The week-old tool beat the six-year incumbent

On August 18, Anthropic replaced its Workbench, the prompt-testing tool in its developer Console, with Playground. Anthropic did more than just change the tool’s name, though.

Anthropic also removed features like saved prompts, version history, evals, and team sharing. The new tool is stateless, meaning it remembers nothing and stores nothing on Anthropic’s servers. (If you have anything stored in Workbench, you have until September 1 to export your old Workbench data.)

Playground may be new to Anthropic, but we’ve seen it before at OpenAI. OpenAI’s Playground is one of the original AI tools. It’s been around since the days of GPT-3. It launched in June of 2020, a full two and a half years before ChatGPT existed. 

Worth noting is that the same week Anthropic launched its leaner Workbench replacement, OpenAI announced it would shut down its saved Prompts and Evals platform on November 30. Both companies reached the same conclusion: Prompts belong in your code, not in a web console.

Both companies reached the same conclusion, prompts belong in your code, not in a web console.

I wanted to see how the newer Playground holds up against OpenAI’s long-standing Playground. Did Anthropic launch a comparable tool, or were they just checking a box by adding a lower-quality product to keep up with their “always be launching” goals? I put both to the test to find out.