PANews reported on May 13 that OpenAI announced the launch of HealthBench, a new evaluation benchmark for AI medical systems, which was designed by 262 doctors from 60 countries and covers 5,000 real simulated conversations. HealthBench uses the scoring criteria set by doctors to test the accuracy, completeness and clinical practicality of the model's response. The code and data set are now open.

In addition, OpenAI announced this morning that all Plus, Team and Pro users can now export in-depth research reports as well-formatted PDF files, including tables, pictures, citations and source links. This feature applies to both new and old reports, and will be available to Enterprise and Edu version users later.