Anthropic launches self-hosted Managed Agents and Claude Fable 5.2 models
New release notes detail infrastructure options for Managed Agents and pricing changes for the latest reasoning-focused model variants.
Key points
- Self-hosted Claude Managed Agents allow users to run the entire service on their own infrastructure.
- Claude Fable 5.2 costs $8 per MTok input and $40 per MTok output, cheaper than previous versions.
- The new models enforce strict prefix checks and drop thinking blocks when switching to other models.
Anthropic added a launch notice for self-hosted Claude Managed Agents to its API release notes on 12 November 2026.reported The update allows customers to operate the complete Managed Agents service, including agent storage and session history, on their own infrastructure rather than using Anthropic-hosted sandboxes.Anthropic now lets companies run its full agent service on their own servers, not just in the cloud.The 12 November 2026 release notes expand self-hosting from sandboxes to the full Managed Agents service stack. The server routes model requests to a chosen provider, with Anthropic serving as the default option. Access requires contacting an Anthropic account team.
Evidence: reported
Claude API Release Notes page updated: We’ve launched self-hosted Claude Managed Agents. You run the whole Managed Agents servicepagewatch-anthropic
How we got here
On 9 November 2026, Anthropic introduced Claude Fable 5.2 and Claude Mythos 5.2.reported These models feature a one-million-token context window and always-on adaptive thinking. Mythos 5.2 is restricted to organisations verified through Anthropic’s programs, such as the Cyber Verification Program.reported Both models cost $8 USD per million input tokens and $40 USD per million output tokens, marking a reduction from the $10 and $50 rates for version 5.1.reported
Evidence: reported
Anthropic says AI is eroding Nvidia's CUDA lead, cutting an AMD chip switch from months to three daysaifleet-decoder
Evidence: reported
Anthropic says AI is eroding Nvidia's CUDA lead, cutting an AMD chip switch from months to three daysaifleet-decoder
What is new
The latest release notes confirm that self-hosted Managed Agents are now available. Previously, self-hosting options were limited to sandboxes. Now, users can store agents and session data on their own infrastructure. The system sends requests to a selected provider, defaulting to Anthropic. This expands the deployment options for enterprise customers seeking greater control over their agent environments.
What to watch
Developers should monitor compatibility issues when switching models mid-conversation. Thinking blocks from Fable 5.2 or Mythos 5.2 are dropped when moving to other models like Opus 5.5 or Fable 5.1.reported The advisor tool in beta only accepts Fable 5.2 or Mythos 5.2 as advisors for 5.2 executors.reported Requests using the older computer_20251124 toolset return errors on most platforms, except Amazon Bedrock.reported
Evidence: reported
Everything you wanted to know about Nvidia but were too afraid to askaifleet-fastcompany
Evidence: reported
Claude API Release Notes page updated: We’ve launched self-hosted Claude Managed Agents. You run the whole Managed Agents servicepagewatch-anthropic
Availability and limits
Self-hosted Managed Agents require direct contact with an Anthropic account team for access. Claude Fable 5.2 is available on the Claude API, Amazon Bedrock, AWS, Google Cloud and Microsoft Foundry.reported Mythos 5.2 requires verification.reported Both models mandate 30-day data retention.reported Zero data retention is unavailable unless expressly authorised. Thinking blocks are only preserved within the originating account or linked accounts. Prefix checks are enforced for all accounts.
Evidence: reported
Claude API Release Notes page updated: We’ve launched self-hosted Claude Managed Agents. You run the whole Managed Agents servicepagewatch-anthropic
Evidence: reported
Anthropic says AI is eroding Nvidia's CUDA lead, cutting an AMD chip switch from months to three daysaifleet-decoder
Evidence: reported
Everything you wanted to know about Nvidia but were too afraid to askaifleet-fastcompany
Why it matters
- Developers can now host full agent services on private infrastructure, storing agents and session history locally.
- The price drop for Fable 5.2 lowers the cost per token compared to the previous 5.1 version.
- Strict prefix enforcement and dropped thinking blocks require code refactoring for applications that switch models dynamically during conversations.
- Beta advisor tool restrictions mean Fable 5.2 and Mythos 5.2 executors accept only 5.2 models as advisors.
Topics: Anthropic, Claude, Managed Agents, Fable 5.2, API
Updated
Then / now / next
Self-hosting options were limited to sandboxes, and Fable 5.1 cost more per token.
Full Managed Agents can be self-hosted, and Fable 5.2 offers lower pricing with stricter compatibility rules.
Developers should test their code for prefix mismatches and advisor tool compatibility before migrating.
Story timeline
- Then Self-hosting options were limited to sandboxes, and Fable 5.1 cost more per token.
- 11 Nov 2026 Anthropic says AI is eroding Nvidia's CUDA lead, cutting an AMD chip switch from months to three days
- 12 Nov 2026 Claude API Release Notes page updated: We’ve launched self-hosted Claude Managed Agents. You run the whole Managed Agents service
- 12 Nov 2026 Google launches Gemini 4 Hydrogen, calling it a major upgrade over Gemini 3.8 Flash, at $1.50/1M input and $7.50/1M output tokens across its 1M context window (Koray Kavukcuoglu/Google)
- 12 Nov 2026 Gemini 4 Hydrogen and Orchestrator
- 12 Nov 2026 Everything you wanted to know about Nvidia but were too afraid to ask
- 12 Nov 2026 Google's Gemini 4 Hydrogen outscores the bigger Argon just four weeks after its launch
- 12 Nov 2026 Catch Go bugs that fuzzing misses with Gemini and a constraint solver
- 12 Nov 2026 Claude Meetings lets Anthropic's AI join your video calls
- 12 Nov 2026 Google’s new Gemini model is better at understanding complicated questions
- 12 Nov 2026 v2.1.323
- 12 Nov 2026 Artificial Analysis' Intelligence Index: Gemini 4 Hydrogen nearly ties Claude Sonnet 5.5 and GPT-6.1 Sol, at a third of Sonnet's cost but ~2.5x that of Sol (@artificialanlys)
- 12 Nov 2026 Google releases Gemini 4 Hydrogen, a lower-cost model said to outscore Argon
- 12 Nov 2026 llm-gemini 0.37
- 13 Nov 2026 Gemini Subscriptions page updated: Project Genie6
- 13 Nov 2026 0.168.0
- 13 Nov 2026 Gemini Subscriptions page updated: Project Genie7
- Now Full Managed Agents can be self-hosted, and Fable 5.2 offers lower pricing with stricter compatibility rules.
- Next Developers should test their code for prefix mismatches and advisor tool compatibility before migrating.
The strongest doubt
Counterpoint from the sources
The self-hosted option requires direct sales engagement, which may delay access for some users.
Numbers in context
Circle area is proportional to the magnitude; labels retain signs. Figures may describe different measures. Read their contexts above.Ask the article
Local keyword search of this article; no generated answers.Did this article give you the evidence you needed?
Your choice stays on this device. No public tally.Questions readers ask
How do I access self-hosted Managed Agents?
You must contact your Anthropic account team to request access to the self-hosted service.
What happens to thinking blocks when I switch from Fable 5.2 to Opus 5.5?
The API drops the thinking blocks, though the request itself still succeeds.
Can I use the older computer toolset with Fable 5.2?
No, requests using computer_20251124 return a 400 error on all platforms except Amazon Bedrock.
Connected stories
Reading receipt
- Self-hosted Claude Managed Agents allow users to run the entire service on their own infrastructure.
- Claude Fable 5.2 costs $8 per MTok input and $40 per MTok output, cheaper than previous versions.
- The new models enforce strict prefix checks and drop thinking blocks when switching to other models.