OpenAI has released its new gpt-oss model, and Google is now supporting its deployment on Google Kubernetes Engine (GKE) with optimized configurations. GKE is designed to manage large-scale AI workloads, offering scalability and performance with advanced infrastructure, including GPU and TPU accelerators. Users can quickly get started with the GKE Inference Quickstart tool, which simplifies the setup and provides benchmarking capabilities.

OpenAI's gpt-oss models utilize the harmony response format to structure conversation outputs, reasoning, and function calls. This format allows for flexible output channels and is designed to integrate seamlessly with existing APIs, while custom implementations can follow the provided guide for proper formatting. Users are encouraged to refer to the documentation for comprehensive instructions on using the format effectively.

Perplexity evaluates OpenAI's newly released open-weight models, gpt-oss-20b and gpt-oss-120b, focusing on their implementation on NVIDIA H200 GPUs. The article discusses infrastructure decisions, kernel modifications, and performance optimizations made to efficiently integrate these models into their inference engine, ROSE.