studioglobal
熱門發現
報告已發布6 來源

Claude Mythos Preview:2026 年 4 月亮相,跑分領先但未一般開放

Claude Mythos Preview 於 2026 年 4 月 7 日隨 Anthropic Project Glasswing 公開亮相;Anthropic 系統卡目錄亦將它列入 April 2026。[12][30] 官方 benchmark 顯示,Claude Mythos Preview 在 GPQA Diamond、Humanity’s Last Exam、BrowseComp、OSWorld Verified 等項目高過 Claude Opus 4.6。[16] 但它並非一般公開產品。Anthropic 表示沒有將 Mythos Preview 作 general availability 發布,而是只向部分...

17K0
抽象 AI 模型与网络安全界面的编辑插画,代表 Claude Mythos Preview 与关键软件安全
Claude Mythos Preview:2026 年 4 月亮相、跑分领先但暂不公开可用AI 生成的 Claude Mythos Preview 主题编辑配图,表现前沿模型与软件安全场景。
AI 提示

Create a landscape editorial hero image for this Studio Global article: Claude Mythos Preview:2026 年 4 月亮相、跑分领先但暂不公开可用. Article summary: Claude Mythos Preview 于 2026 年 4 月 7 日随 Project Glasswing 公开亮相;Anthropic 公布的 GPQA Diamond、Humanity’s Last Exam、BrowseComp、OSWorld Verified 等跑分均高于 Claude Opus 4.6,但它没有 general availability,只向部分网络安全合作伙伴开放。[12][15][16]. Topic tags: ai, anthropic, claude, cybersecurity, software security. Reference image context from search candidates: Reference image 1: visual subject "## InfoQ Software Architects' Newsletter. A monthly overview of things you need to know as an architect or aspiring architect. Live Webinar and Q&A: Architecting for Autonomous Rel" source context "Anthropic Releases Claude Mythos Preview with ... - InfoQ" Reference image 2: visual subject "## InfoQ Software Architects' Newsletter. A monthly overview of things you need to know as an architec

openai.com

先講最重要一點:**Claude Mythos Preview 已經公開披露,但唔等於一般用戶可以喺 Claude 入面自由揀來用。**公開資料顯示,它在 2026 年 4 月隨 Anthropic 的 Project Glasswing 亮相;Anthropic 將 Project Glasswing 定位為用早期前沿 AI 保護關鍵軟件的計劃,並稱該計劃由 Claude Mythos Preview 支援。[12][30] 不過,Anthropic 的系統卡亦講得很清楚:Mythos Preview 並沒有作 general availability 發布。[15]

一句講清楚

  • 發布時間:如果「發布」指公開披露,關鍵日子是 2026 年 4 月 7 日;Anthropic 的模型系統卡目錄亦將 Mythos Preview 列在 April 2026。[12][30]
  • 實力:按 Anthropic 公布的直接對比,Claude Mythos Preview 在 GPQA Diamond、Humanity’s Last Exam、BrowseComp、OSWorld-Verified 等 benchmark 都高過 Claude Opus 4.6。[16]
  • 可用性:它暫時唔係一般公開版 Claude。Anthropic 表示,存取權只提供予若干維護重要軟件基建的合作夥伴組織,並以條款限制在網絡安全用途。[15]

2026 年 4 月 7 日公開亮相,但唔等於正式開放

問「Claude Mythos Preview 幾時發布」,要先分清兩件事。

第一,是公開披露。Anthropic 的 Project Glasswing 頁面標示公告日期為 2026 年 4 月 7 日,並介紹 Project Glasswing 是由 Claude Mythos Preview 支援的關鍵軟件安全計劃。[12] Anthropic 的模型系統卡目錄亦把 Mythos Preview 歸入 April 2026。[30] 所以較準確的講法是:Claude Mythos Preview 在 2026 年 4 月,隨 Project Glasswing 一同公開亮相。

第二,是面向所有用戶的一般開放。這點答案是否定的。Anthropic 的系統卡明確表示,Claude Mythos Preview 沒有作 general availability 發布。[15]

它到底是甚麼?重點係關鍵軟件安全

Project Glasswing 並唔是一個普通聊天機械人產品發布頁。Anthropic 將它描述為一項為 AI 時代保護全球關鍵軟件的計劃,並稱計劃由其「most capable model yet」Claude Mythos Preview 支援。[12]

這個定位亦解釋了為何 Mythos Preview 沒有直接開放給一般用戶。Anthropic 系統卡指出,這類能力既可以用於發現及修補軟件漏洞,也可能被用來設計更複雜的漏洞利用方式;正因如此,Anthropic 決定不將 Mythos Preview 一般公開發布。[15]

Anthropic 紅隊博客的說法亦一致:Mythos Preview 最初透過 Project Glasswing 向有限的關鍵行業合作夥伴及開源開發者開放,目標是在類似能力更廣泛可用之前,先讓防守方加固重要系統。[4]

跑分有幾強?官方數據高過 Claude Opus 4.6

按 Anthropic 在 Glasswing 頁面公布的直接對比,Claude Mythos Preview 在以下多項 benchmark 均高於 Claude Opus 4.6:[16]

BenchmarkClaude Mythos PreviewClaude Opus 4.6
GPQA Diamond94.6%91.3%
Humanity’s Last Exam(without tools)56.8%40.0%
Humanity’s Last Exam(with tools)64.7%53.1%
BrowseComp86.9%83.7%
OSWorld-Verified79.6%72.7%

Anthropic 亦註明,在 BrowseComp 上,Claude Mythos Preview 不單分數較高,而且使用的 token 比 Opus 4.6 少 4.9×。[16]

換句話講,如果你所講的「實力」是指 Anthropic 公開評測中的高難度問答、工具使用、瀏覽任務及電腦環境任務,Mythos Preview 的表現確實非常強。再加上它在 Project Glasswing 的定位,現有公開材料最能支持的結論是:它在複雜推理、工具使用,以及軟件安全相關任務上具備明顯的前沿能力。[12][16]

不過,唔好只睇分數就下定論

這些數字很搶眼,但唔應該簡單推論成「所有真實場景都全面領先」。原因有三個。

首先,本文可引用的直接跑分來自 Anthropic 自身公布的頁面,並非外部獨立復現結果。[16] 這不代表分數不可信,但解讀時應視之為官方披露的評測。

其次,Anthropic 自己對 Humanity’s Last Exam 的高分亦有保留:Mythos 在 low-effort 條件下仍然表現良好,這可能表示存在某程度的記憶化。[16] 因此,HLE 分數可以說明模型在該評測上的表現,但不能單獨證明它在所有未知任務上都會同幅度領先。

第三,風險更新材料顯示,能力與安全表現並非每個細項都一面倒。Anthropic 的 Mythos Preview alignment risk update 提到,在 Minimal-LinuxBench 中,Mythos 的表現會因設定而略高於或略低於 Claude Opus;同一摘要亦提到,在 SHADE-Arena 中,長 chain-of-thought 轉錄條件下的 stealth success rate 由約 0% 增至約 3.8%。[13]

較穩陣的判斷是:Claude Mythos Preview 在 Anthropic 公開的關鍵評測中明顯強於 Claude Opus 4.6;但公開證據仍不足以證明它在日常中文寫作、低延遲互動、價格體驗,或所有一般用戶任務上都全面領先。[13][16]

一般用戶而家用唔用到?

不應把 Claude Mythos Preview 當成已經上線的大眾版 Claude。Anthropic 系統卡寫明,它沒有 general availability;存取權提供給若干維護重要軟件基建的合作夥伴組織,並透過條款將用途限制在網絡安全範圍。[15]

所以,如果你關心的是 Claude.ai 入面可唔可以直接切換、常規 API 是否可調用、價格如何、延遲如何、context window 有幾大,現有公開資料未足以提供完整產品答案。按 Anthropic 已披露資料,Mythos Preview 更像是一個高能力、高敏感度、面向防禦性安全工作的受限前沿模型,而不是一般用戶可以即刻使用的新 Claude 版本。[12][15]

結論

Claude Mythos Preview 的準確定位是:2026 年 4 月隨 Project Glasswing 公開亮相、面向關鍵軟件安全的受限 preview 模型。Anthropic 公布的 benchmark 顯示,它在多項指標上高於 Claude Opus 4.6;但 Anthropic 同時明確表示沒有將其一般公開發布,而是限制予網絡安全相關合作夥伴使用。[12][15][16][30]

對一般用戶而言,重點未必是「要唔要即刻轉用 Mythos」,而是:它目前並唔是一個可以直接選用的常規 Claude 型號。

Studio Global AI

Search, cite, and publish your own answer

Use this topic as a starting point for a fresh source-backed answer, then compare citations before you share it.

使用 Studio Global AI 搜尋並查核事實

重點

  • Claude Mythos Preview 於 2026 年 4 月 7 日隨 Anthropic Project Glasswing 公開亮相;Anthropic 系統卡目錄亦將它列入 April 2026。[12][30]
  • 官方 benchmark 顯示,Claude Mythos Preview 在 GPQA Diamond、Humanity’s Last Exam、BrowseComp、OSWorld Verified 等項目高過 Claude Opus 4.6。[16]
  • 但它並非一般公開產品。Anthropic 表示沒有將 Mythos Preview 作 general availability 發布,而是只向部分維護重要軟件基建的合作夥伴開放,並限制於網絡安全用途。[15]

人們還問

「Claude Mythos Preview:2026 年 4 月亮相,跑分領先但未一般開放」的簡短答案是什麼?

Claude Mythos Preview 於 2026 年 4 月 7 日隨 Anthropic Project Glasswing 公開亮相;Anthropic 系統卡目錄亦將它列入 April 2026。[12][30]

首先要驗證的關鍵點是什麼?

Claude Mythos Preview 於 2026 年 4 月 7 日隨 Anthropic Project Glasswing 公開亮相;Anthropic 系統卡目錄亦將它列入 April 2026。[12][30] 官方 benchmark 顯示,Claude Mythos Preview 在 GPQA Diamond、Humanity’s Last Exam、BrowseComp、OSWorld Verified 等項目高過 Claude Opus 4.6。[16]

接下來在實務上我該做什麼?

但它並非一般公開產品。Anthropic 表示沒有將 Mythos Preview 作 general availability 發布,而是只向部分維護重要軟件基建的合作夥伴開放,並限制於網絡安全用途。[15]

接下來我應該探索哪個相關主題?

繼續“Claude Opus 4.7、GPT-5.5、DeepSeek V4、Kimi K2.6:2026 Benchmark 點睇先唔會睇錯”以獲得另一個角度和額外的引用。

開啟相關頁面

我應該將其與什麼進行比較?

對照「DeepSeek V4 唔止 1M context:MoE、長上下文同 API 遷移重點」交叉檢查此答案。

開啟相關頁面

繼續你的研究

研究對話

研究問題

克劳德·米索斯什么时候发布?实力如何?

Studio Global AI32 來源

引用的答案

Bottom line

如果你说的是 Anthropic 的 Claude Mythos Preview,它的官方公开亮相时间是 2026 年 4 月 7 日,随 Project Glasswing 一起公布;Anthropic 的系统卡目录也把它标成 2026 年 4 月发布。[10][29]
就官方已公开的信息看,它非常强,尤其在高难推理、工具使用、浏览/电脑操作和安全相关软件任务上,公布分数整体高于 Claude Opus 4.6。[23][26]
但它不是面向所有用户的一般可用版,而是受限 preview,只开放给部分合作伙伴,并限制在网络安全场景使用。[23]

Key findings

  • Mythos Preview 是 Anthropic 在 2026 年 4 月公开的受限预览模型,不是大众开放产品。[10][23][29]
  • Anthropic 将它描述为其“most capable model yet”,并把它放在 Project Glasswing 这个“关键软件安全”项目中发布。[10]
  • 官方 benchmark 中,Mythos Preview 对 Opus 4.6 全面领先:GPQA Diamond 94.6% vs 91.3%,Humanity’s Last Exam 无工具 56.8% vs 40.0%,带工具 64.7% vs 53.1%,BrowseComp 86.9% vs 83.7%,OSWorld-Verified 79.6% vs 72.7%。[26]
  • Anthropic 也给了保留意见:Mythos 在 Humanity’s Last Exam 上“低 effort 仍表现很好”可能意味着存在一定记忆化因素,因此高分不应被简单等同于所有真实场景都同幅度领先。[26]

Confirmed facts

  • 官方公告显示,Project Glasswing 于 2026 年 4 月 7 日发布,并说明该项目由 Claude Mythos Preview 提供支持。[10]
  • Anthropic 的系统卡目录把 Mythos Preview 列为 “April 2026”。[29]
  • Anthropic 明确表示没有把 Mythos Preview 做一般公开发布,而是仅向部分维护重要软件基础设施的合作伙伴开放,并限制用途为网络安全。[23]
  • 按 Anthropic 公开分数,Mythos Preview 在多项关键 benchmark 上超过 Claude Opus 4.6。[26]

What remains inference

  • 如果你说的“实力”是日常聊天、写作、中文表现、稳定性、延迟和价格的综合体验,现有公开证据还不够完整;目前最强证据主要集中在高难推理、软件理解、工具调用和安全相关能力。
  • 从它的发布方式看,它更像“高能力但高风险、因此受限”的 frontier preview,而不是成熟的大众产品。

What the evidence suggests

  • 如果只看 Anthropic 官方口径,Mythos Preview 可以视为当时 Claude 体系里非常靠前、甚至最前沿的一档模型。[10][23][26]
  • 它的优势看起来尤其集中在需要深度推理、调用工具、浏览信息、操作计算机环境,以及理解和修改复杂软件的任务上。[23][26]
  • 对普通用户来说,当前更重要的结论不是“能不能用到最强 Claude”,而是“这个模型现在并不面向你直接开放”。[23]

Conflicting evidence or uncertainty

  • 目前直接证据几乎都来自 Anthropic 官方材料,所以“官方内部口径一致”这一点较强,但“外部独立复现已经充分验证”这一点还不够强。[10][23][26]
  • 官方风险更新摘要提到,在某些内部测试里,Mythos 的表现只是略高于或略低于 Claude Opus,取决于设置;这说明它并不是在所有子任务上都稳定大幅领先。[22]
  • 官方自己也承认,部分高分可能受记忆化影响。[26]

Open questions

  • Mythos Preview 何时会从受限 preview 变成一般可用产品,目前我没看到明确时间表。
  • 它的价格、API 可用性、延迟、上下文窗口等面向开发者的完整信息,公开证据还不充分。Insufficient evidence.

Sources worth trusting most

  • Project Glasswing 官方公告:确认首次公开时间和官方定位。[10]
  • Mythos Preview System Card:确认它不是一般公开版,而是受限开放。[23]
  • Glasswing benchmark 页面:看 Mythos 与 Opus 4.6 的直接分数对比和 caveat。[26]
  • Anthropic 系统卡目录:交叉确认发布时间属于 2026 年 4 月。[29]

Recommended next step

如果你愿意,我下一条可以直接帮你做一个“Claude Mythos Preview vs Claude Opus 4.7 vs GPT-5 vs Gemini 2.5 Pro”的简明对比,重点看:

  • 代码能力
  • 推理能力
  • Agent 能力
  • 普通用户是否能用
  • 性价比

Summary

简短说:Claude Mythos Preview 是 2026 年 4 月初公开的受限预览模型,官方公开亮相点是 4 月 7 日。[10][29]
按 Anthropic 公布的数据,它很强,而且在多项 benchmark 上超过了 Claude Opus 4.6。[26]
但它目前不是普通用户随便可用的常规 Claude,而是带使用限制的高能力预览版。[23]

來源

  • [4] Claude Mythos Preview \ red.anthropic.comred.anthropic.com

    But the transitional period may be tumultuous regardless. By releasing this model initially to a limited group of critical industry partners and open source developers with Project Glasswing, we aim to enable defenders to begin securing the most important s...

  • [12] Project Glasswing - Anthropicanthropic.com

    Skip to footer Try Claude Claude Project Glasswing Claude Project Glasswing Claude Project Glasswing An initiative to secure the world’s most critical software with early access to frontier AI Read the announcement Announcements NEW Project Glasswing Apr 7,...

  • [13] [PDF] Alignment Risk Update: Claude Mythos Preview - Anthropicanthropic.com

    We have two “secret side task” evaluations that involve secret-keeping: SHADE-Arena, and Minimal-LinuxBench (the latter is non-public, but we believe it better represents our internal usage distribution). The most notable change in performance compared with...

  • [15] Claude Mythos Preview System Cardanthropic.com

    and fixing vulnerabilities in software code) and offensive purposes (designing sophisticated ways to exploit those vulnerabilities). It is largely due to these capabilities that we have made the decision not to release Claude Mythos Preview for general avai...

  • [16] Project Glasswing: Securing critical software for the AI era - Anthropicanthropic.com

    GPQA Diamond Mythos Preview 94.6% Opus 4.6 91.3% Humanity’s Last Exam Mythos Preview without tools 56.8% Opus 4.6 without tools 40.0% Mythos Preview with tools 64.7% Opus 4.6 with tools 53.1% Humanity’s Last Exam: We have found Mythos still performs well on...

  • [30] Model system cards - Anthropicanthropic.com

    Model Date System card --- Mythos Preview April 2026 Read system card Claude Sonnet 4.6 February 2026 Read system card Claude Opus 4.6 February 2026 Read system card Claude Opus 4.5 November 2025 Read system card Claude Haiku 4.5 October 2025 Read system ca...