In a move hailed by scientists as both daring and democratic, the American concern known as OpenAI has this week released the inner workings — the "weights," as engineers put it — of two of its newest reasoning engines.

These calculating machines, christened GPT-OSS-120B and GPT-OSS-20B, are said to rival the finest electronic minds presently available, and have been offered under what the company terms an "Apache 2.0 licence," meaning they may be studied, adapted, and deployed without fee.

It is the first time in five years that OpenAI has unlatched the door to its intellectual workshop, the last such occurrence being the release of GPT-2 in 2019. In recent years, the firm has jealously guarded its most powerful machines, citing both commercial advantage and the grave risk of misuse.

Now, however, there is a perceptible change in the wind. Observers suggest that growing competition — with rivals such as Meta and DeepSeek unveiling their own "open-source" marvels — has spurred OpenAI to act, lest it be seen as lagging behind.

Professor Harold Smythe of Cambridge, interviewed yesterday, declared:

"It is rather as if Rolls-Royce had handed out blueprints for its jet engines. This generosity of spirit, if genuine, could accelerate research across the Empire and beyond."

The models, say those who have tested them, are adept at answering questions, drafting correspondence, and solving abstruse puzzles. The larger of the pair, GPT-OSS-120B, is particularly nimble in matters of logical reasoning, though both occasionally indulge in what researchers politely call hallucinations — the confident invention of facts.