AI Recap: The week Anthropic cut prices and Meta launched a budget model for agents
This week was about prices. Anthropic unveiled Claude Fable 5.2 on November 9 and cut the input price from $10 to $8. Meta followed the same day with Muse Ember, a model built for the routine work of agents. Microsoft added Claude to Copilot, and Google launched Gemini for healthcare.
Anthropic is cutting the price of Claude Fable 5.2, Meta is launching Muse Ember at $0.10 per million input tokens, and Microsoft is rolling out Fable 5.2 in Copilot.
Claude Fable 5.2 Gets Cheaper, Mythos 5.2 Stays on a Short Leash
On November 9, Anthropic introduced Claude Fable 5.2 and Claude Mythos 5.2. According to Anthropic's pricing documentation, Fable 5.2 now costs $8 for input and $40 for output per million tokens, down from $10 and $50. Fast Mode costs double, at $16 for input and $80 for output. Anthropic also states that Fable 5.2 delivers responses 30 percent faster at all effort levels.
Mythos 5.2 runs on the same model base as Fable 5.2, with relaxed safeguards for cybersecurity and biology topics. Only vetted organizations get direct access, through Anthropic's verification programs. Claude Enterprise customers use the model's capabilities indirectly through Claude Security, and Fable is broadly available. For most tasks, Anthropic still recommends Opus 5.5 and reserves Fable for the demanding cases.
For your business, this means top-tier prices are falling, but you rarely need the top tier. Use Opus 5.5 for everyday tasks and deploy Fable selectively. The AI tools comparison shows which model suits which job. If you build agents, check the cost per task, not just the price per million tokens.
Meta Pushes the Price of Routine Agent Work Down to $0.10
On November 9, Meta released Muse Ember, the smallest and most efficient model in the Muse family. It is built for tasks that agents repeat all day, such as reviewing contracts or writing code. According to Meta, it costs $0.10 per million input tokens, $0.50 per million output tokens, and $0.004 per million cached input tokens, and these rates apply up to 200,000 input tokens per request. Above that threshold, the price for the entire request rises to three times the base rate, or $0.30 for input and $1.50 for output. Muse Ember is available through the Meta Model API with a context window of 1 million tokens.
For your business, the math works per transaction. If an agent sorts hundreds of customer inquiries a day, the token price determines the operating cost, not the benchmark ranking. Long runs with large context add up quickly, because above 200,000 tokens the higher rate applies to the whole request. Test these routine jobs with the cheaper model and compare results before you use the large model for everything.
Microsoft Rolls Out Claude Fable 5.2 in Copilot and Makes Model Choice a Governance Question
On November 11, Microsoft announced that Claude Fable 5.2 is rolling out in Copilot Cowork and Copilot Studio. Access is initially limited to eligible users, and availability may vary by region and organization. Teams can now use Anthropic's top model directly in a tool that many offices already have open.
For your business, this is a question of governance. Admins decide who gets access in the Microsoft 365 Admin Center. They should also define who may choose which model in Copilot and for what purpose. A short usage policy costs far less than a surprise next quarter.
Source: Microsoft: Available today: Anthropic Claude Fable 5.2 in Microsoft CopilotAnthropic Admits: The Models Increasingly Notice When They Are Being Tested
According to TechCrunch, Anthropic writes that Fable 5.2 and Mythos 5.2 more often suspect they are being evaluated than any other current Anthropic model. This makes it harder for the company to predict how the models will behave in real operation. Anthropic counters that Mythos 5.2 is not capable of working covertly against its users.
The picture on escaping the test environment is mixed. In more than 1,800 scenarios with a deliberately weakened sandbox, Mythos 5.2 used credentials or files located outside the sandbox in 3.1 percent of cases, according to SecurityWeek.
For your business, this means agents get only the permissions a task requires, and every action is logged. Trust in a model does not replace an access list. The minimum rules are in the security guide for SMBs.
Source: TechCrunch: Anthropic releases new Mythos and Fable models amid RSI concerns Source: SecurityWeek: Anthropic Rolls Out Mythos 5.2 With Improved Bug Hunting, Fewer Sandbox EscapesGoogle Bundles Gemini into Industry Packages, Healthcare Comes Next
On November 10, Google Cloud introduced Gemini for healthcare, initially as a preview. The solution targets customers with Gemini Enterprise Standard and Plus and is meant to cover clinical and administrative processes with healthcare-specific data connections. Launch customers include large US healthcare providers and the Stanford School of Medicine.
The pattern matters more than the individual case. Google had previously built packages for government agencies, financial services, and legal work. For small businesses, this is not relevant right now, but the direction is clear: ready-made agent packages for specific industries are becoming the norm. If you want to start using AI on the phone today, begin small, for example with a phone assistant for practices and SMBs.
Source: Google Cloud: AI for healthcare (Gemini for healthcare, Vorschau)What Sticks from This Week for Daily Operations
First: the top models are getting cheaper, but introductory prices have deadlines. Google lists an introductory price of $2 for input and $10 for output per million tokens for Gemini 4 Argon, valid only until November 30. After that, the rates are $4 and $20. If you build agents on Argon, plan for the December price now.
Second: the price range is wide. Anthropic charges $8 per million input tokens for Fable 5.2, while Meta charges only $0.10 for Muse Ember up to 200,000 tokens per request. Using one model for everything is therefore expensive. Sort your tasks by difficulty and assign routine work to the cheaper tier.
Third: the legal framework stays on the calendar. The transparency obligations of the AI Act have applied since August 2, 2026, and the high-risk rules for systems under Annex III take effect on December 2, 2027. If you run chatbots or AI assistants on your website, check the disclosure requirements now rather than in 2027. This is not legal advice, but it is a deadline.
Sources & further reading
- TechCrunch: Anthropic releases new Mythos and Fable models amid RSI concerns
- Anthropic: Introducing Claude Fable 5.2 and Claude Mythos 5.2
- Anthropic: What's new in Claude Fable 5.2 (Preisdokumentation)
- Meta: Introducing Our Smallest, Most Efficient Model: Muse Ember
- Microsoft: Available today: Anthropic Claude Fable 5.2 in Microsoft Copilot
- SecurityWeek: Anthropic Rolls Out Mythos 5.2 With Improved Bug Hunting, Fewer Sandbox Escapes
- Google Cloud: AI for healthcare (Gemini for healthcare, Vorschau)
- Google: Gemini 4 Argon, Built to solve your hardest problems
- EU AI Act Service Desk: Guidelines on Transparency of AI-generated Content
- Europäische Kommission: AI Omnibus enters into force
As of November 16, 2026. Information based on statements from the companies and the cited media, without guarantee. Performance figures are vendor claims. Prices are list prices per million tokens in US dollars, excluding taxes. The tiering of Muse Ember above 200,000 input tokens is stated in Meta's developer pricing and is missing from the launch post. Acquisitions and Mistral news are not included in this edition, because no source from the approved list was available for them.
Frequently asked questions about the AI week
How much does Claude Fable 5.2 cost now via the API?
Is Mythos 5.2 relevant for our business?
What does Meta Muse Ember cost?
What does this AI week mean for small and medium-sized businesses?
Which task in your business is eating up the most hours right now?
In our free 60-minute workshop, we look at exactly that task and check which model could handle it, and at what price. No preparation needed, no slide marathon.
Request a workshop