Grok Voice Beats GPT Realtime in VulcanBench with Lower 'Voice Tax'
XFreeze · x · 2026-08-02
In a head-to-head voice model comparison by VulcanBench, xAI's Grok Voice Think Fast 2.0 outperformed OpenAI's GPT Realtime.
Both models were asked the same 200 questions in text and spoken audio formats. The results show:
- Grok: 99.0% text accuracy → 95.7% audio accuracy (-3.3 pp)
- GPT Realtime: 97.5% text → 93.5% audio (-4.0 pp)
Grok not only achieved higher overall accuracy but also had a smaller "voice tax"—the accuracy lost when switching from typed text to spoken audio. Furthermore, Grok completed each turn nearly 2× faster than GPT, although GPT started speaking sooner. This demonstrates that effective voice AI must listen, reason, and solve tasks correctly in real-time, beyond just sounding human.
More from Models
- Google won't publish Argon 4 limits in AI Pro—users compare experiences — Consistent-Text-2662 · 2026-10-16
- Tencent releases Hy-Memory-30B-A3B, a 3B-active memory-focused model claiming to beat Opus — victormustar · 2026-10-16
- Reddit user puzzled why so many prefer Claude over 'much better' ChatGPT — -ButterflyEffect- · 2026-10-16
- User rants GPT-6.1 Sol is painfully slow and buggy, says Claude is 7x faster — karmay007 · 2026-10-16
- $20-plan users have no idea how far AI agents have come in 10 months — realmeetjames · 2026-10-16
- Tencent Hunyuan open-sources Hy Memory 2.0: model, SDK and MemHub for agent memory — TencentHunyuan · 2026-10-16