GPT-4.5’s early system card, which undercut the launch hype — confirmed

GPT-4.5’s launch-day leak was notable for its restraint rather than a spectacular promise. A document circulating ahead of the 27 February 2025 announcement described a large new OpenAI model while warning against assuming it surpassed the company’s reasoning systems in every important area. The Verge recorded that wording and the later edits to the document.

That qualification was quickly overshadowed by the usual launch vocabulary about intelligence and capability. But size, conversational polish and performance on difficult reasoning tasks are different things. A model can improve on its predecessor in one respect while remaining behind another product from the same company in another.

The Verge and TechCrunch reported that OpenAI subsequently removed a statement describing GPT-4.5 as not a frontier model. The edit is part of the record; it is not evidence that the original caution was fabricated. Nor does an early system card independently validate every benchmark or safety claim inside it. Those remain claims made by the organization that conducted the evaluations.

OpenAI’s public launch confirmed that GPT-4.5 existed and was being offered as a research preview. Its announcement emphasized broader knowledge, conversational behavior and continuing limitations. That supports a confirmed badge for the product and early document episode, not a blanket endorsement of the model’s reliability or marketing comparisons.

The leak mattered because it showed the tension between technical qualification and product positioning just before release. Readers were briefly able to see caveats that did not fit neatly into a bigger-is-better narrative. This archive entry uses the public announcement and contemporary outlet reporting, not a hosted copy of the early document. We did not verify a suitable original X post for this story, so it is presented as sourced reporting without an X embed or engagement figures.