xAI has released Grok Voice 3.0, a new speech-to-speech model for its Voice Agent API. Independent benchmarking firm Artificial Analysis lists the model under the vendor name SpaceXAI, and ranks it first on its Speech-to-Speech Index with a score of 86.9. That places it ahead of OpenAI's GPT-Live-1, at 83.2, and Google's Gemini 3.8 Live, at 82.6.
The model targets voice agents handling live customer calls. xAI says it responds in well under a second and keeps the same $0.08-per-minute price as Grok Voice 2.0, undercutting several rival voice APIs.
The index result is not a clean sweep. On blind listening tests, where people chat with each model and pick a favourite, Grok Voice 3.0 comes third, behind two Gemini models. On the τ-Voice benchmark, which simulates support calls, it scores 71.6%, second to GPT-Live-1's 74.5%. So the model leads on one widely-watched composite score while trailing on tests closer to real conversation quality.
Voice agents are becoming a genuine commercial battleground, not just a chatbot feature. Call centres and customer service teams need low latency, reliable turn-taking and a price that scales across thousands of calls a day. xAI's bet is that a top ranking on a public leaderboard, plus a held price point, will pull that business away from OpenAI and Google. Whether blind human preference or leaderboard rank matters more to buyers is still an open question.
Source: MarkTechPost
