Jan 12, 02:40 AM

AI Model 'o1-preview' with 32B Parameters Claims Comparable Performance for $450, Sparking Skepticism Over Benchmark Maxing

Recent discussions among AI experts have surfaced claims regarding a new model, referred to as 'o1-preview', which purportedly achieves performance comparable to existing high-level models while being trained for less than $450. This assertion has raised skepticism within the community, with some experts questioning the validity of the benchmarks and the feasibility of such low training costs. Notably, one commentator expressed disbelief over the possibility of training a reasoning model that meets the benchmarks of o1-preview for such a low price, suggesting that there may be underlying issues with the reported results. Others have speculated on the implications of this development, particularly regarding the capabilities of larger AI labs that typically invest billions in their models.

Written with ChatGPT (GPT-4o mini).

AI Model 'o1-preview' with 32B Parameters Claims Comparable Performance for $450, Sparking Skepticism Over Benchmark Maxing

Sources

Additional media