ElonMuskOur AI startupsxAAnnouncement of the official launchGrok-1.5,The official push didn’t say anything, just threw out the link, with the main message being “less is more”.
What upgrades are there in Grok-1.5? There are two main aspects:
1. Long context understanding
For the context window,Grok-1.5 directly increased it to 16 times the previous level.It increased from 8192 to 128k, which is on par with GPT-4.
This means that Grok-1.5 can handle longer and more complex prompts while maintaining its ability to follow instructions.
In the Needle in an Haystack (NIAH) evaluation, Grok-1.5 demonstrated powerful retrieval capabilities, retrieving embedded text in contexts up to 128K in length and achieving perfect retrieval results.
2. Ability and Reasoning
Grok-1.5maximumOne of the improvements is the ability to handle programming and math-related tasks.It surpasses Grok-1, Mistral Large and Claude 2 in all aspects.
In mathematics, Grok-1.5 scored 50.6% on the MATH benchmark, surpassing the medium-sized Claude 3 Sonnet; and scored 90% on GSM8K.
In terms of programming, Grok-1.5 scored 74.1% on the HumanEval benchmark.It surpasses the medium-sized Claude 3 Sonnet, Gemini Pro1.5, and GPT-4, and is second only to the large-sized Claude 3 Opus.
