SpaceXAI shipped Grok Voice Think Fast 2.0 on July 29, 2026, and the numbers put it ahead of every comparable model on the market. On Artificial Analysis’ speech-to-speech benchmark, Think Fast 2.0 scored 82.9%, beating GPT-Realtime-2.1 at 79.1% and Gemini 3.1 Flash at 69.5%. For businesses deploying voice AI agents at scale, this is the most capable commercially available voice model right now.

The previous version, Think Fast 1.0, scored 75.7% on the same benchmark. That 7-point jump in a few months reflects how quickly the voice AI market is moving, and how quickly the gap is opening between leaders and everyone else.

What Changed

The headline improvements in Think Fast 2.0 are latency, accuracy, and efficiency, in that order.

Latency: Time to first audio dropped from 1.25 seconds to 0.70 seconds. That 0.55-second reduction is the difference between a voice agent that feels like it is thinking and one that feels like it is actually present in the conversation. In customer service applications, that gap shapes whether someone keeps talking or hangs up.

Transcription accuracy: SpaceXAI reports 1.5 to 2 times better transcription compared to Deepgram Nova 3 and ElevenLabs Scribe v2 on their evaluation set. The model is also explicitly designed to handle noisy environments and varied accents, two real-world conditions that trip up most voice AI deployments in practice.

Token efficiency: Think Fast 2.0 uses roughly 60% fewer reasoning tokens than version 1.0 to produce the same output. That matters for cost at scale. A voice agent handling hundreds of thousands of calls per month will see the difference in billing even if the per-minute rate stays flat.

The model supports more than 25 languages, and the architecture is built to listen, reason, and speak simultaneously rather than waiting for turn-taking cues before processing. That concurrent approach is what keeps the conversation flowing without the awkward gaps that make AI voice agents feel robotic.

Pricing is $0.08 per minute of audio. That rate already made Think Fast 1.0 competitive. With the performance improvements in 2.0, it is now both cheaper and more capable than most alternatives.

Built Into Agent Builder

Think Fast 2.0 is available directly on SpaceXAI’s Grok Voice Agent Builder, the no-code platform that lets businesses deploy production-grade phone agents without requiring engineering resources to configure the underlying model.

For existing users, SpaceXAI is handling the migration automatically. The grok-voice-latest model alias flips to Think Fast 2.0 on August 5, 2026. Teams that want to stay on version 1.0 can pin grok-voice-think-fast-1.0 explicitly before that date.

SpaceXAI also noted real-world results from an A/B test run on Starlink’s phone service, where Think Fast 2.0 produced higher sales conversion and support containment rates compared to the previous model. Starlink and SpaceXAI share infrastructure under the SpaceX parent company, so those numbers come from a deployment running at meaningful scale.

What This Means for Business

Voice AI is compressing fast. Six months ago, the realistic options for enterprise voice deployments were either expensive and capable or cheap and limited. That gap has narrowed considerably, and Think Fast 2.0 narrows it further.

The latency benchmark is now below 1 second. Most enterprise voice AI buyers have used 1 second as an informal threshold for conversational quality. Think Fast 2.0 at 0.70 seconds crosses into territory where the model response will feel natural to most callers, not just tolerable. That changes the business case for applications that previously required human agents because AI latency was too visible.

Multi-language support at this performance level is significant. Businesses operating across multiple markets or serving diverse customer bases have historically had to accept lower quality voice AI in non-English languages. Think Fast 2.0’s 25-language support at top-tier benchmark scores changes that calculation.

The pricing sets a floor for the market. At $0.08 per minute, SpaceXAI is pricing aggressively relative to performance. Competitors will respond, which means the cost of enterprise-grade voice AI is coming down regardless of which platform a business chooses.

For any business running or evaluating voice AI for customer service, internal support, or sales applications, the smart move right now is to benchmark Think Fast 2.0 against whatever you are currently using. The performance improvements are significant enough that the comparison is worth doing.

---

Enterprise DNA builds AI voice agents for businesses that want to handle calls, surface knowledge, and automate communication without scaling headcount. If you want to understand what current voice AI can actually do in your business context, start with a discovery call.