tech
December 17, 2025
Announcing Grok-1.5
Grok-1.5 comes with improved reasoning capabilities and a context length of 128,000 tokens. Available on š soon.
TL;DR
- Grok-1.5, the latest model from xAI, offers long context understanding and advanced reasoning.
- It will be available to early testers and existing Grok users on the š platform.
- Grok-1.5 shows improved performance in coding and math, scoring 50.6% on MATH and 90% on GSM8K.
- It achieved a 74.1% score on the HumanEval benchmark for code generation.
- The model can process contexts up to 128K tokens, significantly increasing its memory capacity.
- It demonstrated perfect retrieval capabilities in the Needle In A Haystack evaluation.
- Grok-1.5 is built on a custom distributed training framework using JAX, Rust, and Kubernetes.
- The training infrastructure is designed for reliability and uptime, with automatic node detection and optimized checkpointing.
Continue reading
the original article