OpenAI internal model beats GPT-6 Astra on every BenchCAD point
On the new BenchCAD chart, OpenAI's internal model tops GPT-6 Astra at every published point, from about 0.76 versus 0.594 up to a peak of 0.875, roughly 47% higher.
OpenAI’s internal model/gpt 6.2 already beats GPT‑6 Astra at maximum reasoning at EVERY published point on its new BenchCAD chart. The lowest internal-model point is around 0.76 versus Astra’s 0.594. Its peak reaches 0.875—a roughly 47% higher reported score. That’s the part I keep thinking about: today’s maximum could become tomorrow’s starting point. These tasks involve reconstructing 3D parts, assembling components and understanding circuit connections. If gains like this carry into coding, research and longer agent workflows, we could be looking at the beginning of a new era.
This post quotes another post: “OpenAI shares ChatGPT and Codex progress in hardware engineering”