SpaceXAI launches Grok Voice 3.0, tops Artificial Analysis speech index

SpaceXAI has released Grok Voice 3.0, a new model built to power voice agents. Announced as of 15 October 2026, the model now holds the top spot on the Speech-to-Speech Index maintained by independent benchmarking firm Artificial Analysis, with a score of 86.9.

Grok Voice 3.0 works natively from speech to speech. According to SpaceXAI, it can listen to a user, reason about the request, call external tools when needed and reply out loud, all within a single model.

Aimed at developers

The release is pitched primarily at developers building voice applications. SpaceXAI is offering the model through an endpoint that can be plugged into third-party products. Details such as pricing, supported languages and latency figures were not available in the information released so far.

Tool calling is the feature that lets a voice model do more than talk. It allows an assistant to connect to services such as calendars, search or customer-support systems, turning a conversation into a completed task rather than just an answer.

Why the ranking matters

Artificial Analysis compares AI models on measures including quality, speed and cost. Its Speech-to-Speech Index focuses on systems that take audio in and produce audio out. Reaching first place there gives SpaceXAI a strong performance claim against competing voice models.

Voice has become one of the most contested fronts in AI. Speech-to-speech models skip the traditional chain of transcribing, generating text and synthesising audio, which promises faster and more natural exchanges. Pushing the Grok brand further into voice signals SpaceXAI's ambition to win more of the developer market.

Benchmark leadership does not always translate into real-world quality, however. How Grok Voice 3.0 performs in production deployments will become clearer as developers begin testing it.