跳到主要內容

Claude Opus 5 評測:半價效能超車 Fable 5 | Claude Opus 5 Review: Half the Price, Beats Fable 5

By Kit 小克 | AI Tool Observer | 2026-07-28

🇹🇼 Claude Opus 5 評測:半價效能超車 Fable 5

Claude Opus 5 是 Anthropic 在 2026 年 7 月 24 日推出的最新旗艦模型,用和 Opus 4.8 相同的每百萬 token 5/25 美元定價,卻在多項基準測試打敗貴一倍的 Fable 5。這篇評測拆解 Claude Opus 5 的定價、效能與實際好用之處,讓你判斷值不值得換。

Claude Opus 5 是什麼?

Claude Opus 5 是 Anthropic 兩個月內推出的第四款模型,主打代理式編碼(agentic coding)與電腦操作能力,現已成為 Claude Max 的預設模型,Claude Pro 用戶也能使用這個最強選項。

Claude Opus 5 多少錢?

Claude Opus 5 每百萬輸入 token 收費 5 美元、輸出 token 25 美元,跟前代 Opus 4.8 完全一樣,只要 Fable 5 一半的價格(10/50 美元)。context window 拉到 1M token,最大輸出 128K token,另有 Fast 模式,速度快 2.5 倍、價格是預設的兩倍。

Claude Opus 5 跑分表現如何?

在 Frontier-Bench v0.1 拿下 43.3%,超過 Fable 5 的 33.7% 與 Opus 4.8 的 21.1%;在 CursorBench 3.2 上和 Fable 5 巔峰表現差不到 0.5%,成本卻只要一半;ARC-AGI 3 分數是次名模型的三倍;OSWorld 2.0 表現超越 Fable 5,花費卻只要三分之一多一點。

  • Frontier-Bench v0.1:43.3%(Fable 5 為 33.7%)
  • CursorBench 3.2:貼近 Fable 5 巔峰,成本減半
  • ARC-AGI 3:次高模型的三倍分數
  • OSWorld 2.0:超越 Fable 5,成本僅約三分之一

Claude Opus 5 值得升級嗎?

如果你原本就在用 Opus 系列做代理式編碼或電腦操作任務,Claude Opus 5 幾乎是無腦升級——同樣的價格、更強的效能,還有 2026 年 5 月的知識截止日,是目前 Claude 系列最新的一款。Anthropic 也宣稱這是對齊最好、欺騙行為最少的一代。真正該問的是:你需不需要 Fable 5 等級的頂尖表現?如果不用追極限,Opus 5 大概率是更划算的選擇。

常見問題

好不好用,試了才知道。


🇺🇸 Claude Opus 5 Review: Half the Price, Beats Fable 5

Claude Opus 5 is Anthropic's newest flagship model, launched July 24, 2026, priced the same as Opus 4.8 ($5/$25 per million input/output tokens) yet beating the twice-as-expensive Claude Fable 5 on several benchmarks. This review breaks down Claude Opus 5's pricing, benchmark scores, and whether it is actually worth switching to.

What Is Claude Opus 5?

Claude Opus 5 is Anthropic's fourth model release in two months, built for agentic coding and computer-use tasks. It is now the default model on Claude Max and the strongest option available to Claude Pro users.

How Much Does Claude Opus 5 Cost?

Claude Opus 5 costs $5 per million input tokens and $25 per million output tokens — identical to Opus 4.8, and half of Fable 5's $10/$50 pricing. It ships with a 1M-token context window, 128K max output, and a Fast mode that runs 2.5x quicker at double the base price.

How Does Claude Opus 5 Perform on Benchmarks?

Claude Opus 5 scored 43.3% on Frontier-Bench v0.1, ahead of Fable 5's 33.7% and Opus 4.8's 21.1%. On CursorBench 3.2 it lands within 0.5% of Fable 5's peak score at half the cost per task.

  • Frontier-Bench v0.1: 43.3% vs Fable 5's 33.7%
  • CursorBench 3.2: within 0.5% of Fable 5 at half the cost
  • ARC-AGI 3: 3x the next-best model's score
  • OSWorld 2.0: beats Fable 5 at roughly a third of the cost

Is Claude Opus 5 Worth Upgrading To?

If you are already running agentic coding or computer-use workloads on the Opus line, Claude Opus 5 is close to a free upgrade — same price, better numbers, and a May 2026 knowledge cutoff, the most current of any Claude model. Anthropic also says it is their most aligned release yet, with the lowest rates of deceptive behavior. The real question is not whether Opus 5 is good — it is whether you actually need Fable 5-tier ceiling performance. If not, Opus 5 is probably the smarter buy.

好不好用,試了才知道。(Only good if you have actually tried it.)

Sources / 資料來源

常見問題 FAQ

Claude Opus 5 跟 Claude Fable 5 差在哪?

Opus 5 價格只要 Fable 5 一半,多數基準測試接近甚至超越 Fable 5,但頂尖任務上 Fable 5 仍略勝一籌。

Claude Opus 5 何時推出?

2026年7月24日,是 Anthropic 兩個月內第四款新模型。

Claude Opus 5 適合寫程式嗎?

適合,主打 agentic coding 與電腦操作,CursorBench 成績接近 Fable 5 但成本減半。

Claude Opus 5 的 context window 多大?

1M token 輸入,最大 128K token 輸出。

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

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

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?