All posts
Claude Published Updated 13 min

November 9 agent guide: Claude Fable 5.2 and Muse Ember

November 9, 2026 was a frontier-plus-worker day. Anthropic released Claude Fable 5.2 and Claude Mythos 5.2, Meta shipped Muse Ember for high-volume agent subtasks, and Claude Code 2.1.321 made Fable 5.2 the default Fable model.

Jigar JoshiJigar JoshiAgentic AI Architect and Consultant
November 9, 2026 cover for Claude Fable 5.2 and Muse Ember
In this post (8 sections)

November 9, 2026 is the largest model day in this Bacancy Technology window. Anthropic upgraded Claude Fable and Claude Mythos to 5.2, cut price versus 5.1, and added an optional fast mode. Meta released Muse Ember for the high-volume steps agents repeat all day. Claude Code 2.1.321 wired Fable 5.2 in as the default Fable model. OpenAI also expanded dots so one agent can use multiple local computers.

What shipped on November 9

November 9, 2026 primary developments
ReleaseSourceAccessBuilder action
Claude Fable 5.2 / Mythos 5.2AnthropicFable broad; Mythos verified programsRe-bench hard coding and science tasks; keep Opus 5.5 as daily driver
Muse Ember (muse-ember-1.0)MetaMeta Model API / Muse CodeAdd as subagent and routing worker at $0.10 / $0.50
Claude Code 2.1.321AnthropicChangelogConfirm default Fable points at claude-fable-5-2
OpenAI dots multi-computerOpenAIPro / Business (admin rules vary)Connect a second machine and run one thread per computer

Claude Fable 5.2 and Mythos 5.2

Anthropic positions Fable and Mythos for the hardest open-ended technical work. Fable 5.2 and Mythos 5.2 are the same underlying model. Fable keeps strong cyber and biology safeguards so any customer can use it. Mythos is available through expanded Life Sciences and Cyber Verification Programs. Anthropic reports outputs up to 30% faster than 5.1, clearer communication, and lower list pricing at $8 / $40 per million tokens. Fast mode is optional at $16 / $80.

Vendor-reported capability highlights

  • FrontierCode 2.0: 49.1% at Xhigh effort (Anthropic’s highest released-model score in its table).
  • FrontierSWE v2: 69.9%.
  • AutomationBench: 51.4%.
  • Terminal-Bench-Science 0.1: 74.6% (75.3% Mythos 5.2).
  • OSWorld 2.1 offline partial: 89.3%.
  • Humanity’s Last Exam with tools: 74.2%.

Anthropic recommends starting most coding and knowledge work on Claude Opus 5.5, then moving up to Fable or Mythos for ambitious jobs, or down to Sonnet 5.5 or Haiku 5.5 for cheaper volume. Fable can also act as Advisor or Orchestrator over smaller models. Treat all scores as vendor-reported until independent harnesses land.

Muse Ember: cheap volume for agent loops

Meta’s Muse Ember is the worker model for tool calls, coding subtasks, routing, extraction, and long-document parsing. Meta cites Artificial Analysis efficiency-frontier placement versus Claude Haiku 5.5 and GPT-6 Luna, AutomationBench-AA at 37.8, Agents’ Last Exam at 61.6, and Terminal-Bench 4.0 at 40.9 under max reasoning in Meta’s tables. Meta’s routing line is direct: Muse Wonder plans, Muse Ember handles the volume.

  • 1M-token context and up to 512K output tokens.
  • Native text, image, audio, and video input.
  • Responses API support on Meta Model API.
  • List price $0.10 input / $0.004 cached / $0.50 output per million tokens up to 200K input; $0.30 / $0.012 / $1.50 above that.

Claude Code 2.1.321 and OpenAI dots

Claude Code 2.1.321 adds claude-fable-5-2 as the default Fable model with 1M context and the new $8 / $40 rates, and migrates saved Fable 5.1 picks toward the fable alias when plan policy allows. Separately, OpenAI expanded dots so one persistent agent can connect multiple local computers and choose which machine runs each thread. Keep each machine online with the ChatGPT app open; one thread uses one computer at a time.

Builder checklist

  1. 01
    Re-bench Fable 5.2 on your hardest jobs only
    Keep Opus 5.5 as the default coding model. Promote Fable for migrations, long-horizon science, or hardware tasks where 5.1 or Opus already strained.
  2. 02
    Add Muse Ember as a subagent lane
    Route repetitive tool and coding subtasks to muse-ember-1.0 and measure cost per completed task against Haiku and Luna.
  3. 03
    Confirm Claude Code model defaults
    Upgrade to 2.1.321+, run /model, and verify fleets are not stuck on Fable 5.1 unless intended.

Conclusion

November 9 pairs a cheaper, faster Fable/Mythos frontier tier with Meta’s high-throughput Ember worker. Bacancy Technology’s rule: Opus for daily coding, Fable for the hardest jobs, Ember for volume, Mythos only inside verified programs.

Need a Fable-versus-worker routing matrix?

Bacancy Technology designs model matrices, verification-program readiness, and Claude Code fleet defaults. Consulting: /consulting/ . Contact: /contact/ .

Sources: Anthropic ; research.meta.ai — muse ember ; Claude blog ; learn.chatgpt.com — computers and apps .

The weekly take

Agentic AI patterns, delivered Thursdays

What I am shipping, watching, and pruning out of client stacks each week. One email. No fluff.

Shipping an agentic AI project this quarter?
Book a 30-min consult
Frequently asked

Questions readers ask about this post

Share this post
LinkedIn Facebook WhatsApp