Meet Grok Voice 3.0: SpaceXAI Voice Agent Model Ranks #1 on Artificial Analysis
https://www.marktechpost.com/2026/10/14/meet-grok-voice-3-0-spacexai-voice-agent-model-ranks-1-on-artificial-analysis/📌 Grok Voice 3.0 奪語音測試冠軍，延遲卻不如前代
TL;DR：Grok Voice 3.0 登上 Artificial Analysis 語音榜首，但首音延遲比 2.0 版還慢。
🎣 一款新模型拿下榜首聽起來是全面勝利，但把四個子項分數拆開來看，故事沒那麼單純，SpaceXAI 自己引用的數據裡，就藏著一個退步的指標。
🤔 Speech-to-Speech 榜單登頂
SpaceXAI 發布 Grok Voice 3.0，一款能聽、能推理、能呼叫工具、也能說話的 speech-to-speech 模型，目前在 Artificial Analysis 的 Speech-to-Speech Index 排名第一，得分 86.9。對開發者而言，呼叫端點、事件格式與每分鐘 $0.08 的價格都與 Grok Voice 2.0 相同。模型以 grok-voice-3.0 的名稱掛在 /v1/realtime WebSocket 端點上，而 grok-voice-latest 這個別名已經指向新版，等於既有的 agent 會自動切換到 3.0。SpaceXAI 並未公開參數量或架構細節。預設會開啟 reasoning（reasoning.effort 設為 “high”），也可以設為 “none” 來換取更快的回應速度；barge-in（使用者插話）由伺服器端的語音活動偵測處理；工具呼叫（collections 搜尋、網路搜尋、X 搜尋、遠端 MCP 伺服器、自訂函式）都在對話 session 內執行。
📊 四項子指標與對手的比較
Artificial Analysis 的 Speech-to-Speech Index 是四個子項的等權重平均：Big Bench Audio 推理能力、τ-Voice agentic 表現、Speech Agent Arena 偏好度，以及任務成功率。
| 模型 | AA 綜合分數 |
|---|---|
| Grok Voice 3.0 | 86.9 |
| GPT-Live-1（Astra, medium） | 83.2 |
| Gemini 3.8 Live Extended Thinking | 82.6 |
| Grok Voice 2.0 | 81.3 |
不過子項表現並非全面領先：在 ServiceNow 用來評估企業語音 agent 準確度與體驗的 EVA-Bench 上，SpaceXAI 自己公布的最高準確度分數是 74%。
💡 延遲聲稱「降低 40%」，但其實比前代慢
Artificial Analysis 測得 Grok Voice 3.0 的首音延遲（time to first audio）為 0.81 秒，GPT-Live-1（Astra, medium）為 1.34 秒，SpaceXAI 宣稱「延遲降低 40%」確實成立，但對照自家前代 Grok Voice 2.0 的 0.70 秒，3.0 版反而變慢了。
| 模型 | 首音延遲 |
|---|---|
| Grok Voice 2.0 | 0.70 秒 |
| Grok Voice 3.0 | 0.81 秒 |
| GPT-Live-1（Astra, medium） | 1.34 秒 |
在成本方面，SpaceXAI 引用 Artificial Analysis 的「每個 Big Bench Audio 任務成本」數據：
| 模型 | 每任務成本 |
|---|---|
| Grok Voice 3.0 | $0.025 |
| GPT-Live-1（Astra, medium） | $0.030 |
| GPT-Realtime-2.1（High） | $0.055 |
但若換算成「每小時輸入語音的成本」，Gemini 3.8 Live Extended Thinking 反而更便宜：$3.50，Grok Voice 3.0 則是 $4.80。文中也提醒，各家牌價其實不能直接比較：GPT-Live-1 每分鐘 $0.05，但後端模型與工具使用是另外計費；Gemini 則是語音按分鐘計費外加文字 token 費用。以上 benchmark 數據皆來自 Artificial Analysis，截至 2026 年 10 月 14 日。SpaceXAI 的部落格文章也列出了凝固時間點的 Arena Elo 值：GPT-Live-1 為 1048，Gemini 為 990。
⚠️ 架構黑盒，且榜首地位靠的是綜合分數
SpaceXAI 沒有公開 Grok Voice 3.0 的參數量或架構設計，外界無法得知效能提升的技術來源；而「榜首」頭銜是四項子指標等權重平均後的結果，單看延遲或 EVA-Bench 準確度，Grok Voice 3.0 並非每一項都領先對手。
🎯 實務啟示
API 與 Grok Voice 2.0 相容，且與 OpenAI Realtime API 相容，遷移主要只需換端點網址與金鑰；但正式環境建議直接指定 grok-voice-3.0，而不要依賴會自動變動的 grok-voice-latest 別名，避免模型升級帶來非預期的延遲或行為變化（本次 3.0 版的首音延遲其實比 2.0 版慢就是一例）。Grok Voice 2.0 目前已標示為棄用（deprecated）。
🔗 來源
- 標題：Meet Grok Voice 3.0: SpaceXAI Voice Agent Model Ranks #1 on Artificial Analysis
- 作者／機構：Michal Sutter, MarkTechPost
- 連結：https://www.marktechpost.com/2026/10/14/meet-grok-voice-3-0-spacexai-voice-agent-model-ranks-1-on-artificial-analysis/
#GrokVoice #SpaceXAI #SpeechToSpeech #VoiceAI #RealtimeAPI #ArtificialAnalysis #AIBenchmark #ConversationalAI #MCP #AIAgent
原始資料 MarkTechPost · 收集於 2026-10-15
摘要原文
SpaceXAI just released Grok Voice 3.0 . Grok Voice 3.0 is a SpaceXAI speech-to-speech model that listens, reasons, calls tools and speaks. It now ranks first on the Artificial Analysis Speech-to-Speech Index with a score of 86.9. For developers, the endpoint, events and $0.08 per minute price are unchanged from Grok Voice 2.0. Grok Voice 3.0 is SpaceXAI’s flagship realtime voice model, exposed as grok-voice-3.0 on the /v1/realtime WebSocket endpoint . The grok-voice-latest alias already points to it, so existing agents switched automatically. SpaceXAI has not published parameter counts or architecture details. Reasoning is on by default ( reasoning.effort: "high" ) and can be set to "none" for faster turns. Server-side voice activity detection handles barge-in. Tools run inside the session: collections search, web search, X search, remote MCP servers and custom functions. The AA Speech-to-Speech Index is an equal-weighted average of 4 components. Those are Big Bench Audio reasoning, τ-Voice agentic performance, Speech Agent Arena preference and task success rate. Grok Voice 3.0 scores 86.9. GPT-Live-1 (Astra, medium) scores 83.2, Gemini 3.8 Live Extended Thinking 82.6. Grok Voice 2.0 scored 81.3. Component results are strong, not uniformly first: ServiceNow’s EVA-Bench scores enterprise voice agents on accuracy and experience. SpaceXAI reports a top accuracy score of 74%. Artificial Analysis measures 0.81 s time to first audio. GPT-Live-1 (Astra, medium) measures 1.34 s, so SpaceXAI’s “40% lower latency” claim checks out. Grok Voice 2.0 was faster, though, at 0.70 s. On cost, SpaceXAI cites Artificial Analysis cost per Big Bench Audio task. Grok Voice 3.0 costs $0.025 per task, versus $0.030 for GPT-Live-1 (Astra, medium) and $0.055 for GPT-Realtime-2.1 (High). On AA’s cost per hour of input audio, Gemini 3.8 Live Extended Thinking is cheaper: $3.50 versus $4.80. List prices are not directly comparable. GPT-Live-1 costs $0.05 per minute, but backend model and tool usage is billed separately. Gemini 3.8 Live bills audio per minute plus text tokens. Benchmark data: Artificial Analysis , accessed October 14, 2026. SpaceXAI’s blog lists frozen Arena Elo values of 1048 (GPT-Live-1) and 990 (Gemini). The API is compatible with the OpenAI Realtime API , so migration mostly means a new URL and key. Production teams should pin grok-voice-3.0 rather than the alias. Grok Voice 2.0 is now deprecated. Check out the official announcement , the Speech to Speech docs and the Artificial Analysis leaderboard . All credit goes to the researcher of this project. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter . Wait! are you on telegram? now you can join us on telegram as well. The post Meet Grok Voice 3.0: SpaceXAI Voice Agent Model Ranks #1 on Artificial Analysis appeared first on MarkTechPost .
由 claude-code/sonnet 自動生成