xAI Releases Grok Voice 2.0, Tops Voice Agent Benchmarks

xAI officially released its next-generation end-to-end speech model, Grok Voice Think Fast 2.0, on July 30, making it available via API and Voice Agent Builder. Designed for voice agents in real-world physical environments, the model delivers comprehensive upgrades in noise-robust transcription, complex workflow reasoning, and natural conversation. According to Artificial Analysis, it ranks second in the Speech-to-Speech index (82.9%) with a speech reasoning score of 97.2%, and tops the τ-voice Bench agentic benchmark with 56.5%. First-token latency is reduced to 0.7 seconds, and it supports parallel processing ('thinking while listening'). Pricing is set at $0.08 per minute ($4.80 per hour of input audio), up from v1.0's $3.00/hour but roughly half the price of comparable real-time speech models like GPT.

Confirmed

• Noise robustness and transcription: Outperforms specialized models in accuracy across 24 languages, with 10x noise robustness.
• Benchmarks: Per Artificial Analysis, Grok Voice Think Fast 2.0 High ranks second in the Speech-to-Speech index (82.9%) with a speech reasoning score of 97.2%, and tops τ-voice Bench with 56.5%.
• Latency and parallel processing: First-token latency of 0.7s, supports thinking-while-listening and multitasking.
• Pricing and availability: Available via API and Voice Agent Builder at $0.08/min ($4.80/hour), higher than v1.0 but cheaper than GPT.

Unconfirmed

• Developer testimonials (e.g., CNVS test in m13) praise cost, speed, and intelligence, but specifics are not public; m19 notes that actual conversation effects 'seemed to make testers feel...' without clarifying positive or negative.

Why it matters

• The model directly targets voice interaction in complex real-world environments, solving hearing challenges in noisy settings and lowering cost barriers for building responsive voice agents.
• Artificial Analysis also launched a dedicated Speech-to-Speech leaderboard, providing a multi-dimensional reference for evaluating native speech models.

2026-07-30 ~ 2026-08-01 · 20 related posts

• [source] xAI Launches Grok Voice 2.0: #2 in Speech-to-Speech, 0.70s Latency — ArtificialAnlys · 2026-07-30
• Grok Voice 2.0 Tops Tau Voice Benchmark with Massive Leaps in Full Duplex Scores — ArtificialAnlys · 2026-07-30
• Grok Voice 2.0 Priced at $4.80/Hour: Significantly Cheaper Than GPT — ArtificialAnlys · 2026-07-30
• Artificial Analysis Launches Comprehensive Speech-to-Speech AI Leaderboard — ArtificialAnlys · 2026-07-30
• [source] xAI Officially Announces Grok Voice Think Fast 2.0 — SpaceXAI · 2026-07-30
• [source] xAI Launches Grok Voice 2.0: Noise Robustness and Transcription Accuracy 10x Better than Dedicated Models — SpaceXAI · 2026-07-30
• Grok Voice 2.0 Hits API at $0.08/min with Upgraded Noise Robustness and Reasoning — SpaceXAI · 2026-07-30
• Grok Voice Think Fast 2.0 Released: Major Leaps in Voice Reasoning and Noise Robustness — XFreeze · 2026-07-30
• Grok Voice Think Fast 2.0 tops Artificial Analysis τ-Voice benchmark, excelling in real customer-service tasks — XFreeze · 2026-07-30
• xAI Announces Grok Voice Think Fast 2.0 with Improved Intelligence — kevinnbass · 2026-07-30
• Grok Voice 2.0 Details: 0.7s Latency, 60% Fewer Tokens at $0.08/min — mark_k · 2026-07-30
• xAI's Grok Voice Think Fast 2.0 Tops Speech-to-Speech Benchmark — elonmusk · 2026-07-30
• xAI Launches Grok Voice Think Fast 2.0 — Polymarket · 2026-07-30
• xAI Drops Grok Voice Think Fast 2.0: Major Upgrades in Speed & Agent Handling — Scobleizer · 2026-07-30
• Grok Voice Think Fast 2.0 Claims Top Spot at $0.08 Per Minute — ns123abc · 2026-07-30
• Grok Voice Think Fast 2.0 Released: Fast and Expressive Voice Model — stefanjblos · 2026-08-01

4 near-duplicate retellings: ArtificialAnlys · ArtificialAnlys · XFreeze · 智东西