Elon Musk's xAI Unveils Grok-1.5 with Improved Reasoning Capabilities, 128K Context Window

Elon Musk’s xAI has announced Grok-1.5, which comes with improved reasoning capabilities and a context length of 128,000 tokens. It will be available to our early testers and existing Grok users on the 𝕏 platform in the coming days.

Grok-1.5 boasts notable improvements, particularly in coding and math-related tasks. It beats Mistral Large on various benchmarks including MMLU, GSM8K and HumanEval. 

During tests, Grok-1.5 demonstrated exceptional performance, achieving a remarkable 50.6% score on the MATH benchmark and an impressive 90% score on the GSM8K benchmark. These benchmarks cover a wide range of math problems, showcasing Grok-1.5's versatility and problem-solving capabilities.

In addition to its prowess in math-related tasks, Grok-1.5 excelled in the HumanEval benchmark, scoring 74.1%. This benchmark evaluates code generation and problem-solving abilities, further highlighting Grok-1.5's comprehensive skill set.

A standout feature of Grok-1.5 is its long context understanding capability, enabling it to process contexts of up to 128K tokens within its window. This significant enhancement represents a sixteen-fold increase in memory capacity compared to previous models, allowing Grok-1.5 to utilize information from substantially longer documents.

The Grok-1.5 infrastructure is built on a custom distributed training framework based on JAX, Rust, and Kubernetes. This robust training stack ensures reliability and uptime of training jobs, minimising downtime and maximizing efficiency during large-scale model training.

xAI recently released the model weights and network architecture of Grok-1. As the model is gradually rolled out to a wider audience, xAI plans to introduce several new features to Grok 1.5 in the coming days.