MIT Technology Review reports OpenAI’s case for GPT-4.5—and the doubts it faces

MIT Technology Review reports OpenAI’s case for GPT-4.5—and the doubts it faces

MIT Technology Review published this report on February 27, 2025. It describes OpenAI’s release of GPT-4.5 as a bet that scaling up a general-purpose model can improve its grasp of conversational cues and its performance on some tests. OpenAI researchers Mia Glaese and Nick Ryder describe its conversations as more natural and responsive to implicit expectations; OpenAI testers preferred it to GPT-4o for everyday, professional, and creative tasks. The report also draws a boundary around those claims: GPT-4.5 did better on SimpleQA, but its advantage was smaller on other tests, and it scored worse than o3-mini on standard science and math benchmarks. The benchmark results are not the same as evidence of reliability in everyday use. Writer cofounder and CTO Waseem Alshikh argues that the release is a smoother version of the same approach, with energy costs that may outweigh differences many users notice. Ryder, by contrast, remains optimistic that scaling can continue. The dispute is over whether larger models’ conversational improvements justify the compute and energy they require, or whether efficiency and niche problem-solving deserve more attention. The report does not establish how much users notice the difference or what the energy costs are.

Publication: 2e65f7c1-cccb-947d-becd-a06658e57e3f

attribution — pass: This is a single MIT Technology Review article, not a listing. The reading attributes product claims to OpenAI and its researchers Glaese and Ryder, and the criticism to Writer CTO Waseem Alshikh. The Deep Dive sidebar is not used.

The company claims it is its biggest and best model for all-round chat yet.

summary — uncertain: The central comparison is hard for a general reader to follow. SimpleQA, o3-mini and GPT-4o are named without explanation. A reader cannot tell that o3-mini is a step-by-step reasoning model, which is why its lead on math matters, or that SimpleQA is a general-knowledge quiz. Smallest repair: briefly gloss SimpleQA and o3-mini using the source's own descriptions, or add them to the term library. The summary also leaves out OpenAI's claim of fewer hallucinations.

Unlike reasoning models such as o1 and o3, which work through answers step by step, most large language models like GPT-4.5 spit out the first response they come up with.

date — pass: February 27, 2025 matches originalPublished at day precision and the URL path. There is no hindsight framing. Rollout plans and GPT-5 plans are not presented as outcomes. The sidebar items that refer to later events are not used. This is a publisher page, so no archive disclosure is required.

OpenAI says it will begin rolling out to other users next week.

support — pass: The claims match the source. These include the OpenAI release, its biggest-and-best claim, the Glaese and Ryder quotes, the SimpleQA lead and the weaker science and math results versus o3-mini. Alshikh's energy critique and Ryder's optimism are also supported. The preference findings are correctly limited to testers employed by OpenAI. The open question is framed as an unresolved issue.

Human testers employed by OpenAI say they preferred GPT-4.5 to GPT-4o for everyday queries, professional queries, and creative tasks

media — pass: No media was selected and no verified candidates were supplied, so nothing was misidentified.