Elon MuskMusk's AI startupsxAAnnounced to launchGrok-1.5, the model is able to understand long context andadvanced functionsInference will be open to X platform Grok users and early testers in the near future.
Grok-1.5 islatestthe model is able to understand long context andadvanced functionsInference capabilities. It will be available soon on the 𝕏 platform for early testers and existing Grok users.
Grok-1.5 has significant improvements over previous versions, most notably in coding and math-related tasks. In the math benchmark, Grok-1.5 scored 50.6%, and in the GSM8K benchmark, it achieved 90%. Additionally, it scored 74.1% in the HumanEval benchmark, which evaluates code generation and problem-solving capabilities.
A new feature of Grok-1.5 is the ability to handle contexts up to 128K tokens long. This enables Grok to have up to 16 times the context length of previous versions, allowing it to utilize information from longer documents. Additionally, the model can handle longer and more complex prompts while still maintaining its ability to follow instructions as the context window grows.
InlatestIn the latest Grok-1.5 version, a custom distributed training framework based on JAX, Rust and Kubernetes is also used.