OpenAI May Launch More Capable Internal AI Model as Soon as Friday

OpenAI may release new artificial intelligence models as early as Friday, according to a report by The Information, with speculation centering on an unreleased internal model that substantially outperforms the company's current GPT-6 Astra system on engineering-focused tests. OpenAI has not confirmed any launch date or product details.

The company published benchmark results on November 3 showing that its internal model achieved an average score of 87.5% on BenchCAD 2.0, a test designed to measure AI performance on engineering tasks involving 3D models, assemblies and circuit boards. By comparison, GPT-6 Astra scored 59.4% at its maximum reasoning setting.

A separate post on X claimed the internal model would launch on Friday and deliver major gains over both GPT-6 Astra and GPT-6.1 Sol. The user, @imjustnewatai, shared a detailed comparison of scores and estimated costs across reasoning settings, noting that the internal model's lowest estimated setting scores roughly 76%, still well above Astra's maximum of 59%.

The same comparison put Astra's cost at $0.44 per task for a 14% score, rising to $11.05 per task at its maximum setting. The internal model's lowest setting was estimated at about $11 per task for a score of around 76%, suggesting the performance jump comes with a sharply higher price. The user said the cost figures were extrapolated from OpenAI's benchmark graph and that the reasoning-setting labels were estimates.

The reported development points to a widening focus on specialised, high-cost AI systems aimed at engineering and design work rather than general consumer chat. For enterprise users, the trade-off between accuracy and per-task expense is likely to be a central consideration if such a model reaches the market.

OpenAI has so far declined to comment on the reported launch timeline, and none of the details beyond the published benchmark figures have been independently verified.