跳到主要內容

Gemini 3.6 Flash評測:輸出降58%成本效能反贏旗艦Pro | Gemini 3.6 Flash Review: Cheaper, Faster, Beats Pro on Coding

By Kit 小克 | AI Tool Observer | 2026-08-13

🇹🇼 Gemini 3.6 Flash評測:輸出降58%成本效能反贏旗艦Pro

Gemini 3.6 Flash是Google在2026年7月21日發布的新一代輕量模型,主打「便宜又能打」:輸出token價格比上一代3.5 Flash便宜,官方公布的程式編寫與代理人(agentic)測試分數卻反過來超越自家更貴的旗艦Gemini 3.1 Pro。對常常在API帳單上肉痛的開發者來說,這是近期最值得關注的AI模型更新之一。

Gemini 3.6 Flash是什麼?

Gemini 3.6 Flash是Google Gemini家族的中階快速模型,維持100萬token超長上下文、支援文字/圖片/影片/音訊/PDF多模態輸入,知識截止日更新到2026年3月,比3.5 Flash的2025年1月新了超過一年。

定價怎麼算?

輸入token每百萬字元1.5美元,輸出token每百萬7.5美元,比3.5 Flash的9美元輸出價再降。若用快取輸入,每百萬只要0.15美元,等於直接砍90%。對長對話、長文件處理的應用來說省下的成本相當可觀。

效能真的贏過Gemini 3.1 Pro嗎?

是的,至少在Google公布的官方測試上如此。Gemini 3.6 Flash在DeepSWE程式修復測試進步12個百分點,MLE-Bench機器學習工程測試進步14.2個百分點,代理人多步驟任務也用更少的推理步驟、更少的工具呼叫就完成,程式碼修改精準度更高、較少誤動到不該碰的檔案。跑速也比3.1 Pro快上近一倍。

內建電腦操作(Computer Use)代理人怎麼用?

這代最大亮點是電腦操作功能直接內建為Gemini API和Gemini Enterprise的客戶端工具,不用額外接第三方套件,AI代理人就能直接操作圖形介面完成點擊、輸入、滾動等動作,等於把「AI幫你操作電腦」的門檻降低了一截。

小克怎麼看:值得升級嗎?

如果你原本用Gemini 3.5 Flash做程式助手或自動化代理人,這次升級幾乎沒有理由不換——價格更低、速度更快、代理人任務表現更好。但如果你在乎的是純推理深度而非速度成本,Gemini 3.1 Pro作為旗艦模型還是有它的位置。實際跑幾個你自己的工作流程比對輸出品質,比看官方跑分準。

常見問題

Q: Gemini 3.6 Flash比Gemini 3.5 Flash貴還是便宜?
A: 更便宜,輸出token從每百萬9美元降到7.5美元,快取輸入更是砍到0.15美元。

Q: Gemini 3.6 Flash支援多長的上下文?
A: 維持100萬token輸入上下文,輸出最多可達65,536 token。

Q: 一般開發者現在能用嗎?
A: 可以,已經在Gemini API和Gemini Enterprise正式開放,不是預覽版。

好不好用,試了才知道。


🇺🇸 Gemini 3.6 Flash Review: Cheaper, Faster, Beats Pro on Coding

Gemini 3.6 Flash is Google's new lightweight model launched on July 21, 2026, and its pitch is simple: cheaper output pricing than its predecessor, yet Google's own published benchmarks show it beating the pricier flagship Gemini 3.1 Pro on coding and agentic tasks. For developers watching their API bills, this is one of the more practical model releases in recent weeks.

What Is Gemini 3.6 Flash?

Gemini 3.6 Flash is the mid-tier fast model in Google's Gemini lineup, keeping the 1-million-token context window and multimodal input support for text, images, video, audio, and PDFs. Its knowledge cutoff moved to March 2026, over a year newer than 3.5 Flash's January 2025 cutoff.

How Much Does It Cost?

Input tokens run $1.50 per million and output tokens $7.50 per million — down from $9.00 output pricing on 3.5 Flash. Cached input drops to just $0.15 per million, a 90% discount that matters a lot for long-context or long-conversation apps.

Does It Really Beat Gemini 3.1 Pro?

According to Google's own published numbers, yes. Gemini 3.6 Flash gained 12 percentage points on the DeepSWE code-repair benchmark and 14.2 points on MLE-Bench for ML engineering tasks. It also completes agentic, multi-step workflows using fewer reasoning steps and fewer tool calls, with more precise code edits and fewer unintended file changes — while running roughly twice as fast as 3.1 Pro.

How Does Built-In Computer Use Work?

The headline feature this round is native computer-use support, shipped as a client-side tool directly in the Gemini API and Gemini Enterprise. No third-party plumbing required — an agent can now click, type, and scroll through a GUI on its own, lowering the barrier to building real "AI operates your computer" workflows.

Kit's Take: Worth Upgrading?

If you're already running coding assistants or automation agents on Gemini 3.5 Flash, there's almost no reason not to switch — it's cheaper, faster, and scores better on agentic benchmarks. If raw reasoning depth matters more to you than cost or speed, Gemini 3.1 Pro still has its place. Run your own workflows through both and compare output quality — that beats trusting official benchmarks alone.

FAQ

Q: Is Gemini 3.6 Flash cheaper than 3.5 Flash?
A: Yes — output pricing dropped from $9 to $7.50 per million tokens, and cached input costs just $0.15 per million.

Q: What's the context window?
A: It keeps the 1-million-token input context, with up to 65,536 output tokens.

Q: Is it generally available now?
A: Yes, it's live in the Gemini API and Gemini Enterprise — not a preview.

好不好用,試了才知道。

Sources / 資料來源

常見問題 FAQ

Gemini 3.6 Flash比Gemini 3.5 Flash貴還是便宜?

更便宜,輸出token從每百萬9美元降到7.5美元,快取輸入砍到0.15美元。

Gemini 3.6 Flash支援多長的上下文?

維持100萬token輸入上下文,輸出最多可達65,536 token。

一般開發者現在能用嗎?

可以,已經在Gemini API和Gemini Enterprise正式開放,不是預覽版。

延伸閱讀 / Related Articles


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

這個網誌中的熱門文章

Google Ironwood TPU v7 推理專用晶片解析:效能追平 NVIDIA、成本低 44%,AI 晶片戰爭正式開打 | Google Ironwood TPU v7 Explained: Matching NVIDIA Performance at 44% Lower Cost — The AI Chip War Heats Up

Claude Code 實測:AI 幫你寫程式到底行不行? | Claude Code Review: Can AI Really Code for You?

Cursor vs GitHub Copilot vs Claude Code:AI 程式助手大比拼 | AI Coding Assistants Compared: Cursor vs GitHub Copilot vs Claude Code