What Is Meta Muse?Meta Muse 是什么？
Meta Muse (Muse Wonder / Muse Wonder Full) is the AI agent line released by Meta Superintelligence Labs on October 12, 2026. Where most agents compete on raw context windows, Meta Muse's bet is on memory and focus: Active Memory gives the agent long-term recall across sessions, and Active Focus manages its attention over long-horizon work — paired with strong coding ability for long-range agent tasks.Meta Muse（Muse Wonder / Muse Wonder Full）是 Meta Superintelligence Labs 2026 年 10 月 12 日发布的 AI Agent 产品线。当大多数 Agent 还在卷上下文窗口时，Meta Muse 押注的是「记忆 + 专注」：Active Memory 让 Agent 拥有跨会话的长期记忆，Active Focus 为长程任务做专注力管理，再配上强编程能力，专攻长程 Agent 任务。
It ships in two tiers. Muse Wonder is served through the Meta Model API: $4 input / $20 output per million tokens, $0.20 cache reads, a 1M-token context, and a closed-weights model. Muse Wonder Full is the full tier reserved for Muse Code subscribers — Power $50/month, Maximum $200/month — with no public API pricing. Prices are per the official site.分为两档：Muse Wonder 走 Meta Model API：$4 输入 / $20 输出每百万 token、缓存读取 $0.20、100 万上下文，权重闭源。Muse Wonder Full 是完整档，仅限 Muse Code 订阅用户——Power $50/月、Maximum $200/月，无公开 API 定价。价格以官网为准。
Meta Muse is Meta Superintelligence Labs' attempt to differentiate on the axis everyone else under-invests in: memory. A 1M context window without memory still forgets everything between sessions — Active Memory attacks exactly that gap, and Active Focus is the less glamorous but equally important half (long-horizon tasks die from attention drift, not from lack of context). The honest tradeoffs: it is a closed model with a young ecosystem — the tooling, skills libraries, and community playbooks that mature agents enjoy are still being built — and the Wonder Full tier is gated behind Muse Code subscriptions at $50–$200/month, a steep floor. API pricing at $4/$20 puts it head-to-head with flagship-tier models; the $0.20 cache reads are the real cost lever for memory-heavy, long-running usage. Verdict: if your pain is agents that "forget the plan" across sessions, this is the most memory-native option on the market; if you want the deepest existing agent ecosystem today, the incumbents still lead.
Meta Muse 是 Meta Superintelligence Labs 在一条别人投入不足的轴上做差异化：记忆。没有记忆的 100 万上下文，跨会话照样忘光——Active Memory 正好打这个缺口，而 Active Focus 是不那么光鲜但同等重要的一半（长任务死于注意力漂移，而不是缺上下文）。真实代价：闭源模型、生态年轻——成熟 Agent 享有的工具链、技能库、社区打法都还在建设中；Wonder Full 档被锁在 Muse Code 订阅（$50–$200/月）后面，门槛不低。API 定价 $4/$20 直接对标旗舰级模型，真正省钱的是 $0.20 的缓存读取——对记忆密集、长运行的使用场景很关键。结论：如果你的痛点是 Agent「跨会话就忘计划」，这是市面上记忆最原生的选择；如果要的是今天最深的 Agent 生态，老玩家依然领先。
— 中国吧 AI Tools Hub Editorial Team
Who Should Use Meta Muse?谁适合使用Meta Muse？
✓ Good Fit For适合以下场景
- Teams running long-horizon agent tasks that span sessions and need persistent memory跑跨会话长程 Agent 任务、需要持久记忆的团队
- Meta Model API users who want flagship-tier coding ability at $4/$20 per MTokMeta Model API 用户，想要旗舰级编程能力（$4/$20 每百万 token）
- Muse Code subscribers (Power/Maximum) getting Wonder Full bundledMuse Code 订阅用户（Power/Maximum），直接用捆绑的 Wonder Full
✕ Not Ideal For不适合以下场景
- Self-hosting or open-weights users — Muse is closed-source想自托管/要开放权重——Muse 是闭源模型
- Budget users — Wonder Full starts at $50/month via Muse Code预算敏感用户——Wonder Full 经 Muse Code 起步 $50/月
Getting Started with Meta MuseMeta Muse 快速开始
1. Pick your route: Meta Model API (Muse Wonder, pay per token) or a Muse Code subscription (Power $50/mo / Maximum $200/mo for Wonder Full)选接入方式：Meta Model API（Muse Wonder，按 token 计费）或 Muse Code 订阅（Power $50/月、Maximum $200/月，用 Wonder Full）2. Get your API credentials (or sign in to Muse Code), then point the agent at your project拿到 API 凭证（或登录 Muse Code），把 Agent 指向你的项目3. Let Active Memory learn your conventions once — then hand it long-horizon tasks and let Active Focus keep it on track让 Active Memory 一次记住项目约定，然后把长程任务交给它，Active Focus 会帮它保持专注Key Features核心功能
-
Active Memory: Long-Term RecallActive Memory：长期记忆 — The agent remembers project conventions, decisions, and your preferences across sessions — no more re-briefing every time you start a new task.Agent 跨会话记住项目约定、历史决策和你的偏好，开新任务不用再从头交代一遍。
-
Active Focus: Attention ManagementActive Focus：专注力管理 — Keeps long-horizon agent tasks on track — deciding what deserves the model's focus as work stretches over hours, not minutes.让长程 Agent 任务不跑偏：任务拉长到以小时计时，它依然知道该把注意力放在哪里。
-
Strong Coding Ability强编程能力 — Built for long-range agent tasks: reads codebases, edits files, runs tests, and carries multi-step implementation plans through to completion.为长程任务而生：读代码库、改文件、跑测试，把多步骤的实施计划真正干完。
-
Meta Model API PricingMeta Model API 定价 — Muse Wonder via the Meta Model API: $4 input / $20 output per million tokens, cache reads at just $0.20, 1M-token context. Prices per official site.Muse Wonder 走 Meta Model API：$4 输入 / $20 输出每百万 token，缓存读取仅 $0.20，100 万上下文。价格以官网为准。
-
Muse Code Subscription TierMuse Code 订阅档 — Muse Wonder Full is exclusive to Muse Code subscribers — Power $50/month, Maximum $200/month — with no public API pricing. Prices per official site.Muse Wonder Full 仅限 Muse Code 订阅用户：Power $50/月、Maximum $200/月，无公开 API 定价。价格以官网为准。
Pros & Cons优缺点
✓ Good Fit For适合以下场景
- Active Memory: the most memory-native agent on the market for cross-session workActive Memory：跨会话场景下市面上记忆最原生的 Agent
- Active Focus keeps long-horizon tasks from driftingActive Focus 让长程任务不漂移
- 1M context + strong coding ability in one package100 万上下文 + 强编程能力，一套搞定
- $0.20 cache reads make memory-heavy, long-running usage affordable缓存读取仅 $0.20，记忆密集的长运行场景也负担得起
✕ Not Ideal For不适合以下场景
- Closed model — no self-hosting, no weight downloads闭源模型，无法自托管、无权重下载
- Wonder Full gated behind Muse Code ($50–$200/month)Wonder Full 锁在 Muse Code 订阅后（$50–$200/月）
- Young ecosystem — tooling and community playbooks still maturing生态年轻，工具链与社区打法还在成熟中
Use Cases应用场景
Popular ways people use Meta Muse.人们使用Meta Muse的常见场景：
🧭 Long-Horizon Agent Tasks长程 Agent 任务
Multi-step projects that take hours: Active Focus keeps the plan on track while Active Memory recalls earlier decisions.耗时数小时的多步骤项目：Active Focus 让计划不跑偏，Active Memory 记住此前的决策。
📚 Persistent Project Knowledge持久的项目知识
Teach conventions, standards, and gotchas once — the agent carries them across every future session.约定、规范、坑点教一次，之后每个会话它都记得。
🔧 Codebase Implementation代码库级实现
Describe the feature, let the agent read the repo, implement, test, and iterate with full context.描述需求，Agent 读仓库、实现、跑测试、迭代，全程带着完整上下文。
🔌 Meta Model API IntegrationMeta Model API 集成
Call Muse Wonder at $4/$20 per MTok with $0.20 cache reads — suitable for memory-heavy, long-running product integrations; prices per official site.以 $4/$20 每百万 token、缓存读取 $0.20 调用 Muse Wonder，适合记忆密集的长运行产品集成；价格以官网为准。
Similar AI Tools相似 AI 工具
If Meta Muse doesn't fit your needs, here are other popular AI tools you might consider — see also our full review: OpenAI Codex.如果Meta Muse不符合你的需求，可以看看这些热门 AI 工具——另可参考本站完整评测：OpenAI Codex。