OpenAI's new GPT-5.4 clobbers humans on pro-level work in tests - by 83%

GPT-5.4 is also more reliable, producing 18% fewer errors and 33% fewer false claims than GPT-5.2, according to OpenAI.
Grit Daily

AI Is Writing Your Code, Here’s Why It Needs Its Own QA Layer

TestSprite 2.1 embeds agentic testing into every pull request, catching what AI coding tools miss before bad code ships to ...