xAI tests new object highlighting feature in Grok voice mode

xAI appears to be testing a new "Objects Highlighting" toggle within Grok’s voice mode. This suggests Grok users might soon access a Vision mode where the assistant can visually identify and highlight objects during a conversation. This feature closely resembles the object highlighting announced for Google’s Gemini Live, a capability rooted in the Project Astra demo. Such alignment indicates that xAI is actively tracking developments from other AI labs and working to keep Grok competitive on core multimodal features.

This update is most likely to benefit users who rely on Grok for hands-free, real-world tasks where quickly pinpointing items in an environment is valuable—think accessibility scenarios or smart device control. The technical leap needed here is non-trivial: object detection, visual annotation, and seamless voice integration typically require a major model upgrade. Internally, xAI has been working on “V7” models for improved image and video understanding, with the recently rumored Grok 4.20 potentially being the bridge to these new vision capabilities. Given Grok 4 was based on V6, there’s speculation that 4.20 will represent the next logical step, though an exact launch window remains unknown.

On the Companions front, a switch from voice to pure text chat is in the works, letting users toggle modes and review voice conversation history in the same view. This change mainly streamlines workflow for those using Grok’s assistant-style Companions, offering a clearer division between voice and text exchanges. The actual scope of this rollout is still uncertain, as some text chat functions already exist but may get a dedicated UI for better clarity.

All this comes as xAI recently open-sourced Grok 2.5, with Elon Musk hinting that Grok 3 could also become open within six months. These moves position xAI as a transparent player and show a continued push to expand Grok’s technical reach while responding to rapid advances from rivals like Google and OpenAI.