← 返回首页
RT Mario Nawfal: Grok 4.20 Beta dropped insane benchmark numbers: lowest hallucination rate ever recorded at 22%, number 1 in following instructions a...
RT Mario Nawfal
Grok 4.20 Beta dropped insane benchmark numbers: lowest hallucination rate ever recorded at 22%, number 1 in following instructions at 83%, and #2 in agentic tool use at 97%.
This is a 500B model that was specifically built to tell the truth first, and it’s still crushing the things that actually matter most.
Source: @XFreeze
Elon Musk: Cool, well Grok will get even better every week!
Grok 4.20 Beta dropped insane benchmark numbers: lowest hallucination rate ever recorded at 22%, number 1 in following instructions at 83%, and #2 in agentic tool use at 97%.
This is a 500B model that was specifically built to tell the truth first, and it’s still crushing the things that actually matter most.
Source: @XFreeze
Elon Musk: Cool, well Grok will get even better every week!