SAN FRANCISCO — xAI is moving fast to turn Grok into a full developer platform, rolling out a cluster of upgrades this month that push the model deeper into voice, transcription, and autonomous task execution.
The headline additions give builders more raw capability and more automation. A new speech-to-text API reached general availability, a long-running autonomous mode arrived in Grok Build, and the recently launched Voice Agent Builder rounds out a stack aimed at teams that want to ship production AI features without stitching together half a dozen vendors.
Speech-to-Text Goes Generally Available
The xAI Speech-to-Text API is now generally available, transcribing audio to text across 25 languages with both batch and streaming modes. Streaming support matters for real-time applications like live captioning and voice agents, while batch mode handles large archives of recorded audio, giving developers a single transcription layer that plugs directly into the rest of the Grok toolchain.
It is a natural complement to xAI's voice work, which recently expanded with the no-code Voice Agent Builder for Grok. That platform bundles telephony, knowledge retrieval, tools, guardrails, and observability into one interface and lets an operator stand up a working voice agent in about two minutes, billed at roughly $0.05 per minute of audio.
Grok Build Learns to Run on Its Own
On the coding side, xAI introduced a '/goal' mode in Grok Build, a long-running autonomous capability for handing off larger implementation tasks. Instead of babysitting each step, a developer can hand Grok a goal and let it plan the work, execute until completion, and verify the result, with status, pause, resume, and clear controls to steer the agent along the way.
The autonomous mode builds on Grok Build's expanding developer footprint, including its recent integration with Railway sandboxes for running coding agents. Together, the pieces point toward a workflow where Grok handles multi-step engineering jobs end to end rather than one prompt at a time.
A Voice Model Built to Reason While It Talks
Underpinning the voice tools is Grok Voice, xAI's speech-to-speech model designed to reason in the background so it can work through tricky requests without adding response latency. On the τ-voice benchmark, which tests agents against noisy, real-world calls in dozens of languages, xAI reports its flagship voice model topping the leaderboard at 67.3 percent, ahead of competing real-time offerings.
The steady stream of releases reflects a deliberate strategy: make Grok the place developers build, whether they are transcribing audio, deploying voice agents, or automating software work. Details on each release are posted to xAI's news page. With speech-to-text now generally available and Grok Build running autonomously, xAI is stitching its models into a single, tightly coupled platform, and the pace of shipping suggests there is plenty more on the way.
