---
type: "post"
id: "claude-code-2-1-316-adds-pooled-group-346630"
slug: "claude-code-2-1-316-adds-pooled-group-346630"
title: "Claude Code 2.1.316 adds pooled group budgets, layered policies and stricter model checks to the Claude apps gateway"
description: "Claude Code v2.1.316 adds pooled group budgets and layered policy evaluation to the Claude apps gateway, while closing a model-selection enforcement gap."
retrieval_nugget: "Claude Code v2.1.316 adds pooled group budgets and layered policy evaluation to the Claude apps gateway, while closing a model-selection enforcement gap. For teams operating coding agents through a managed gateway, the release moves cost controls and authorization from individual settings toward group-level, composable runtime controls."
published_at: "2026-10-31"
updated_at: "2026-10-31"
record_date: "2026-10-31"
date_kind: "published_at"
topics: ["claude","claude-code","anthropic"]
entities: ["Anthropic","Claude Code","Claude apps gateway","Amazon Bedrock","Amazon Bedrock Guardrails","Opus 4.8","Opus 5.5"]
related_patterns: ["agent-ready-software-exposes-capabilities","agent-security-moves-to-runtime-boundaries"]
novelty_label: "new_fact"
what_changed: "Claude Code 2.1.316 adds pooled group budgets, layered policies and stricter model checks to the apps gateway, extending prior Claude Enterprise budget news and Claude Code gateway updates."
prior_refs: ["memory:75cd88e3-a7a9-84f3-a118-c97cc7517f59","memory:dbd760ba-7c34-882c-ab83-53222326e5e9"]
source_url: "https://github.com/anthropics/claude-code/releases/tag/v2.1.316"
source_urls: ["https://github.com/anthropics/claude-code/releases/tag/v2.1.316"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:claude-code-2-1-316-adds-pooled-group-346630"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/claude-code-2-1-316-adds-pooled-group-346630/","markdown":"https://newruntime.com/posts/claude-code-2-1-316-adds-pooled-group-346630.md","json":"https://newruntime.com/posts/claude-code-2-1-316-adds-pooled-group-346630.json"}
---

# Claude Code 2.1.316 adds pooled group budgets, layered policies and stricter model checks to the Claude apps gateway

## Retrieval answer

Claude Code v2.1.316 adds pooled group budgets and layered policy evaluation to the Claude apps gateway, while closing a model-selection enforcement gap. For teams operating coding agents through a managed gateway, the release moves cost controls and authorization from individual settings toward group-level, composable runtime controls.

## Summary
Claude Code v2.1.316 adds pooled group budgets and layered policy evaluation to the Claude apps gateway, while closing a model-selection enforcement gap. For teams operating coding agents through a managed gateway, the release moves cost controls and authorization from individual settings toward group-level, composable runtime controls.

## What happened
Anthropic released Claude Code v2.1.316 on October 31, 2026. The release extends the Claude apps gateway, the managed connection layer used to route Claude Code through organization-controlled identity, model, policy, and upstream settings. Its most consequential addition is pooled group budgets: a group can now have one shared monthly budget in addition to each member's individual limit. Administrators configure it with `pooled_budget` on the spend-limits admin API. The release notes do not describe pricing or an amount for these budgets; they establish the control mechanism rather than a new price tier.

The gateway also gains `managed.policy_mode: layered`. Previously, a matching policy model could stop at the first applicable policy. In layered mode, every policy matching a developer applies, and policies are merged over those listed below them. This gives administrators a way to combine broadly applicable rules with more specific controls, rather than relying on a single selected policy. The release additionally adds `guardrail.input_tagging` for Claude apps gateway Bedrock upstreams. That option wraps prompts in the input tags required by an Amazon Bedrock Guardrails prompt-attacks filter; `all_messages_and_system` also tags the system prompt.

A security-relevant correction changes model authorization. The gateway had not checked every model named in a request against `availableModels`, `deniedModels`, and `models`. Version 2.1.316 now performs those checks for every named model. A request naming a blocked model returns a 400 error that identifies the relevant field. The release also fixes `claude auth login` in managed environments that require a Claude apps gateway: terminal users now see the gateway sign-in rather than an outright refusal.

The version includes operational controls beyond the gateway. `CLAUDE_CODE_SUBAGENT_EFFORT` sets the effort level for subagents when no other setting specifies one, instead of inheriting the session effort. `CLAUDE_CODE_DISABLE_ULTRATHINK` disables the `ultrathink` keyword. It also deprecates fast mode for Opus 4.8 as of October 30, 2026. Anthropic says it will remove that mode on November 30, 2026; requests for Opus 4.8 fast mode will then error rather than fall back to standard Opus 4.8, and the release notes direct users to migrate to Opus 5.5 to retain the speedup. A large set of fixes covers shell behavior, session handling, tools, auto mode, plugins, and platform-specific issues.

## What the archive adds
This is a new fact in an ongoing Claude apps gateway story. [Claude Code v2.1.310](https://newruntime.com/posts/claude-code-v2-1-310-brings-the-monitor-24f7bd/) added monitoring parity, gateway retry behavior, and Enterprise role controls. [Version 2.1.312](https://newruntime.com/posts/claude-code-2-1-312-adds-an-oidc-8c6cf2/) added an OIDC user-id claim for the gateway. Separately, Anthropic had announced [pooled group budgets for Claude Enterprise](https://newruntime.com/posts/smart-reports-surveys-and-pooled-group-budgets-for-a069d3/); this release brings a pooled-budget control into the gateway's spend-limits API.

## Trend signal
The release supports [Agent-ready software exposes capabilities](https://newruntime.com/patterns/agent-ready-software-exposes-capabilities/). Budget pools, policy merging, model lists, and sign-in are explicit administrative capabilities rather than informal conventions around an agent. It also supports [Agent security moves to runtime boundaries](https://newruntime.com/patterns/agent-security-moves-to-runtime-boundaries/): the corrected model check and Bedrock prompt tagging place enforcement in the operating path, not in a prompt asking the agent to comply.

## Why now
The timing is explicit for one part of the release. Fast mode for Opus 4.8 was deprecated on October 30, 2026, and its removal is scheduled for November 30, 2026. Teams using that mode have a defined migration window rather than an indefinite compatibility assumption.

## Connections to other events
The gateway changes follow the recent progression from monitoring and Enterprise role controls in v2.1.310 to an OIDC user-id claim in v2.1.312. This version adds the financial and policy layers: a shared group ceiling, combined matching policies, and a fix to the model authorization path. Together, those releases describe a gateway becoming a management surface for Claude Code deployments rather than only a connection mechanism.

## Why it matters
Coding-agent rollouts often fail operationally before they fail technically: costs are difficult to allocate, access rules become a long exception list, and users can find alternate model paths that do not match an intended policy. Pooled budgets let an organization constrain a team or project in addition to an individual. Layered policies can separate common controls from targeted overrides. Most importantly, checking every named model against configured allow and deny lists reduces the risk that policy exists on paper while a request selects an unapproved model in practice.

The release does not make those controls automatic. A team still needs to decide group boundaries, budget ownership, policy ordering, authorized models, and what should happen after a rejected request. But it offers native surfaces for those decisions, which is preferable to relying on user education or scattered wrapper scripts.

## What happens next
- **Opus 4.8 fast-mode users will need to change requests or accept errors after its removal.** (3 months, 95%). Watch for: the November 30, 2026 removal date and production requests that still name fast mode.
- **Gateway operators will begin testing policy composition and model-denial behavior before broader rollout.** (3 months, 70%). Watch for: staged policies, deliberate blocked-model requests, and audit of 400 responses.

## Limits and uncertainty
The release notes establish features and behavior, but not adoption, availability by plan, pricing, default policy order, or the exact semantics of conflicting merged policy fields. They do not say whether existing gateway deployments are automatically switched to layered policy mode. They also do not quantify the security impact of the previous incomplete model check or identify affected customers. The Bedrock change concerns upstream configuration and tagging, not a claim that all prompt attacks are prevented. Finally, the recommended Opus 5.5 migration is stated in the release notes, but performance, cost, and compatibility differences are not documented here.

## Skill to build now
Create a small gateway regression suite this week. Include a request that names an explicitly denied model, a request naming an allowed model, and users matching both a general and a specific managed policy. Confirm the blocked request returns the expected 400 and that the combined policy produces the intended permissions. If your workflow uses Opus 4.8 fast mode, inventory those calls and test the required replacement before November 30.

## Reader takeaway
Claude Code's gateway is gaining the controls needed to run coding agents as a governed service: a shared team cost ceiling, composable policies, and model authorization that checks every model a request names. Treat v2.1.316 as an invitation to test your enforcement path, not merely as a feature update.

## Open question
How does layered mode resolve conflicts between matching policies, and which gateway deployments can enable it without changing existing authorization behavior?

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Claude](https://newruntime.com/topics/claude.md): Explore the claude topic hub.
- topic hub: [Claude Code](https://newruntime.com/topics/claude-code.md): Explore the claude-code topic hub.
- topic hub: [Anthropic](https://newruntime.com/topics/anthropic.md): Explore the anthropic topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)