Gemini 3.8 Live評測:即時語音變聰明,但免費版用不到 | Gemini 3.8 Live Review: Smarter Voice AI, Not Free
By Kit 小克 | AI Tool Observer | 2026-09-18
🇹🇼 Gemini 3.8 Live評測:即時語音變聰明,但免費版用不到
Gemini 3.8 Live 是 Google 在 2026 年 9 月 15 日推出的新一代即時語音 AI 模型,主打「邊想邊講」:一般語音助理是先想完才回答,Gemini 3.8 Live 可以一邊組織答案一邊開口,中間還會自然插入「讓我確認一下…」這類過場話,讓等待感變小。這篇評測直接講結論:技術方向對,但免費用戶目前碰不到,定價也不算便宜。
Gemini 3.8 Live 是什麼?跟一般語音助理差在哪?
Gemini 3.8 Live 是針對低延遲語音對話設計的模型,能即時處理視覺輸入,並在對話中自動偵測、切換 97 種語言,不用手動選語言。另外還有一個進階版本 Gemini 3.8 Live Extended Thinking,專門處理需要多步驟推理的複雜任務,例如技術故障排除、跨 API 訂票,一邊講話一邊在背景跑流程,同步用語音回報進度。
「邊想邊講」實際體驗如何?
這是 Gemini 3.8 Live 最大的賣點。傳統語音助理常見的空白等待,在 Extended Thinking 模式下被拆成一句句過場話填補,體感反應變快。不過這比較像是「隱藏延遲」而不是「消除延遲」——背後運算時間沒少,只是聽起來比較不無聊。
Gemini 3.8 Live 怎麼用?免費版能用嗎?
目前 Gemini 3.8 Live 已經開放給 Gemini API、Google AI Studio 使用,並逐步整合進 Search Live、Workspace(包含 Gmail、Keep)。但要注意:在 Gemini App 裡,同代的 3.8 系列模型僅限 Google AI Pro / Ultra 訂閱用戶使用,免費版目前停留在舊一代的 3.6 Flash,一般使用者短時間內感受不到差異。
定價貴在哪?語音輸出是重點
- 文字輸入:每百萬 token 約 $0.75
- 文字輸出(含思考 token):每百萬 token 約 $4.50
- 語音輸入:每百萬 token 約 $3
- 語音輸出:每百萬 token 約 $12,是文字輸出的近 3 倍
- 不支援 context caching、Batch API,長對話或大量呼叫成本會直接往上疊
值得升級嗎?小克的實測心得
如果你在做的是客服語音機器人、即時口譯這類「延遲敏感」應用,Gemini 3.8 Live 的邊想邊講體驗確實有感,值得認真評估。但如果只是想在手機上跟 AI 聊天、免費體驗,目前還輪不到你——3.8 系列被鎖在付費訂閱後面,加上語音輸出成本不低,個人開發者拿來做小專案要先算清楚帳。Gemini 3.8 Live 的技術展示很漂亮,但「能用」跟「划算」是兩回事。
好不好用,試了才知道。
🇺🇸 Gemini 3.8 Live Review: Smarter Voice AI, Not Free
Gemini 3.8 Live is Google's new real-time voice AI model, launched on September 15, 2026. Its headline feature is "reasoning while talking" — instead of the usual think-then-answer pattern, it starts speaking while still organizing its answer, filling gaps with natural phrases like "let me check that…" to cut perceived latency. Short version: the direction is right, but free-tier users can't touch it yet, and pricing isn't cheap.
What Is Gemini 3.8 Live and How Is It Different?
Gemini 3.8 Live is built for low-latency voice conversation. It processes visual input in near real time and automatically detects and switches between 97 languages mid-conversation, no manual toggling required. Alongside it, Gemini 3.8 Live Extended Thinking handles multi-step reasoning tasks like technical diagnostics or multi-API travel booking, narrating background progress out loud while it works.
Does "Reasoning While Talking" Actually Feel Faster?
This is the core pitch. Extended Thinking mode fills the classic voice-assistant dead air with verbal filler, so it feels more responsive. But it's worth being honest: this is latency-hiding, not latency-removal — the underlying compute time hasn't shrunk, it's just less boring to wait through.
Where Can You Use Gemini 3.8 Live, and Is It Free?
Gemini 3.8 Live is live now through the Gemini API and Google AI Studio, rolling out to Search Live and Workspace (Gmail, Keep). The catch: inside the Gemini app, the 3.8 model family is restricted to Google AI Pro / Ultra subscribers — free-tier users are still on the older 3.6 Flash and won't notice a difference anytime soon.
Pricing: Voice Output Is Where It Gets Expensive
- Text input: ~$0.75 per million tokens
- Text output (including thinking tokens): ~$4.50 per million tokens
- Audio input: ~$3 per million tokens
- Audio output: ~$12 per million tokens — nearly 3x text output
- No context caching or Batch API support, so long conversations or high-volume calls add up fast
Is It Worth Upgrading? Kit's Honest Take
If you're building latency-sensitive products — voice customer support, real-time interpretation — Gemini 3.8 Live's reasoning-while-talking flow is a genuine improvement worth evaluating seriously. But if you just want to chat with AI for free on your phone, it's not your turn yet: the 3.8 family sits behind a paid subscription, and voice output pricing means indie developers should run the numbers before building anything. The demo is impressive; "usable" and "worth the price" are two different questions.
好不好用,試了才知道。
Sources / 資料來源
- Google Official: Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking
- 9to5Google: Gemini 3.8 Live Extended Thinking powers Gemini Live, Gmail, & Keep
- AIReiter: Gemini 3.8 Live API Pricing - What Is Official (2026)
常見問題 FAQ
Gemini 3.8 Live 免費版能用嗎?
目前不行。Gemini App 裡 3.8 系列僅限 Google AI Pro / Ultra 訂閱用戶,免費版仍是舊一代 3.6 Flash。
Gemini 3.8 Live 跟 Extended Thinking 差在哪?
Gemini 3.8 Live 主打低延遲即時對話;Extended Thinking 是進階版,專門處理需要多步驟推理的複雜任務,並邊做邊用語音回報進度。
Gemini 3.8 Live 的語音功能貴嗎?
語音輸出約每百萬 token 12 美元,是文字輸出的近 3 倍,且不支援 context caching 或 Batch API,長時間對話成本會快速累積。
Gemini 3.8 Live 支援多少種語言即時切換?
可自動偵測並在對話中切換 97 種語言,不需要使用者手動設定。
延伸閱讀 / Related Articles
- Firefox Smart Window評測:Mistral進駐,免費AI瀏覽器登場 | Firefox Smart Window Review: Mistral Powers Free AI Browser
- ChatGPT廣告評測:Sponsored Agents開賣,免費版才看得到 | ChatGPT Ads Review: Sponsored Agents Hit Free Tier Only
- Jev評測:TypeSafe新模型號稱444倍省成本,實測剩8.6倍 | Jev Review: TypeSafe's 444x Cost Claim vs 8.6x Reality
AI 工具觀察站 — 每日精選 AI Agent 與工具趨勢
AI Tool Observer — Daily curated AI Agent & tool trends
留言
張貼留言