TechFlow Logo
Login/ Sign up
ETH Gas
Gwei
Fear
gas
7x24hNews

OpenAI employees publicly accuse Grok3's benchmark results of being misleading

2025.02.23 - 03:26
Share

TechFlow news, according to Jinshi News, recently an OpenAI employee publicly accused Elon Musk's xAI company of presenting misleading benchmark results for its latest AI model, Grok3.

In response, Igor Babushkin, co-founder of xAI, insisted the company had done nothing improper.

xAI's chart shows that two versions of Grok3—Grok3 Reasoning Beta and Grok3 mini Reasoning—outperformed OpenAI's current strongest available model, o3-mini-high, on AIME 2025.

However, an OpenAI employee quickly pointed out on X that xAI's chart did not include the AIME 2025 score of o3-mini-high under the "cons@64" condition. Babushkin defended on X, arguing that OpenAI has also previously released similar misleading benchmark charts, although those charts were used to compare its own models' performance.

Add to Favorites
Share to Social Media