xAI Unveils Priority Processing, Context Compaction, and Image Search
xAI rolls out several major API updates, including priority scheduling for inference, a context compaction tool for long conversations, and image search capabilities within Web Search. The updates also bring deeper integration between the Files API and Imagine endpoints.
xAI introduces a priority processing option that allows developers to request higher scheduling priority for text, image, and video inference endpoints by setting a specific service tier parameter. The system only bills users at priority rates when the feature is actively used, and the applied tier is clearly reported in the response metadata. Additionally, a new Public URLs feature enables developers to turn any file stored in the Files API into a shareable, permanent link that can automatically expire or be revoked at any time.
The Files API now integrates seamlessly with the Imagine endpoints to streamline media generation workflows. Developers can directly reference stored files as inputs for image generation without needing to make the files public or re-upload the data. Furthermore, users can configure Imagine requests to automatically save generated assets directly to their Files storage and instantly publish them with a shareable link in a single API round trip.
Several other notable tools arrive to enhance performance and capabilities across the platform. The new Context Compaction API shrinks lengthy conversations into shorter contexts, which reduces costs and speeds up response times for long agent loops. Meanwhile, a WebSocket mode for the Responses API lowers latency for tool-heavy workloads, Smart Turn detection improves the streaming Speech to Text API by reducing false pauses, and Web Search now supports explicit image searching to return relevant visual embeds.