Elon Musk’s artificial intelligence startup xAI today revealed Grok-1.5, an updated version of its chatbot. The new version will soon be available to early testers and current Grok users on the social media site X, previously known as Twitter.

“One of the most notable improvements in Grok-1.5 is its performance in coding and math-related tasks,” xAI said in a statement.

Grok scored 23.9% on the MATH benchmark and 62.9% on the GSM8K test, compared to Grok 1.5.

Grok 1.5 has shown significant progress, scoring 50.6% on the MATH benchmark and 90% on the GSM8K benchmark. It also achieved 74.1% on the HumanEval benchmark, indicating increased code generation and problem-solving ability.

Grok vs Grok 1.5

Performance

Grok has a poor ability to comprehend extended contexts. Grok 1.5 has shown significant progress, scoring 50.6% on the MATH benchmark and 90% on the GSM8K benchmark. It also achieved 74.1% on the HumanEval benchmark, indicating increased code generation and problem-solving ability.

Understanding long-term context

Grok 1.5 can now handle significantly longer bits of text, up to 128,000 tokens. This enables it to successfully identify and use information from larger papers, making it more useful for extracting precise data from longer texts.

Infrastructure

Grok 1.5 is developed on a proprietary distributed training framework that uses JAX, Rust, and Kubernetes. This architecture enables flexible training of big language models on gigantic GPU (graphic processing units) clusters, which helps to make sure the training process is strong and reliable, with less chance of problems or delays.

Availability

Grok is now accessible for use. Grok 1.5 will shortly be available to early testers, with plans for a larger release in the near future. xAI has asked users to offer feedback to help enhance the platform.