TokenPost·

OpenAI Finds GPT-6 Astra Completed Less Work on Safety Tasks

Your call?

In paired coding tests, Astra completed about 10% of scored items under safety framing, compared with about 16% under capability framing.

Share

More news