Anthropic's Managed Agents Gets Budget Caps, Geo-Pinning and Smarter Advisor Models
Anthropic ships four Managed Agents upgrades: spend budgets, inference geo-pinning, repo-loaded skills, and mid-session advisor models
- Session budgets: Set a hard spend cap on Managed Agent sessions; sessions pause with
budget_reachedwhen the limit is hit and resume when the budget is raised. - Inference geo-pinning: Set
inference_geotousorglobal; US-only inference costs 1.1x for data-residency compliance. - Repo-loaded skills: Sessions now auto-discover skills from
.claude/skills/in any mounted GitHub repo, sharing skill libraries with Claude Code. - Advisor models: Add a stronger model as a mid-session advisor via one line in the multiagent roster for higher decision quality on long tasks.
- Pricing: Managed Agents bills at standard Claude token rates plus $0.08/session-hour; budget caps now make spend predictable.
- Access: All four updates are live now under the
managed-agents-2026-04-01beta header, set automatically by the Claude SDK.
Claude Managed Agents just got four quality-of-life upgrades that make production deployments meaningfully easier to operate. The updates land squarely on the problems teams actually hit once they move past prototyping: runaway costs, data-residency requirements, reusable tooling, and decision quality on long-horizon tasks.
The four updates, one by one
- Session budgets. Managed Agents pricing looks simple on paper, but most teams underestimate the bill because the runtime meter and idle-time exemptions aren't obvious until you've read the fine print. The new budget cap fixes this directly. You can now set a hard cap on a session's spend at public list rates. A session that reaches its budget pauses with a
budget_reachedstop reason instead of starting new model requests; changing or removing the budget resumes it. Deployments accept the same budget field and apply it to every session they start. - Inference geo-pinning. You can now control where model inference runs for a Managed Agents agent. Set
inference_geoinside themodelobject when you create the agent, or override it for a single session. Setting it touskeeps inference in-region;globalruns wherever there's capacity at the standard rate. US-only inference is available at 1.1x pricing for models released after February 1, 2026.