BerriAI · yuneng-berri · Aug 10, 2026 · Aug 10, 2026
diff --git a/release_notes/index.md b/release_notes/index.md
@@ -10,18 +10,19 @@ LiteLLM ships new releases regularly with new provider support, performance impr
 
 ## Latest Release
 
-### [v1.95.0 — Claude Opus 5, MCP Gateway DCR & Rust /v1/messages](/release_notes/v1.95.0/v1-95-0)
+### [v1.96.0 — MCP Entitlements, Redis Config Sync & Auto-Router Context](/release_notes/v1.96.0/v1-96-0)
 
-_August 1, 2026_
+_August 9, 2026_
 
-Day-0 support for the 1M-context Claude Opus 5 across Anthropic, Amazon Bedrock, Google Vertex AI, and Azure AI Foundry, joined by Gemini 3.6 Flash and Gemini 3.5 Flash Lite; an MCP gateway front door built on always-on aggregate DCR discovery, identity-only session tokens, RFC 8707 resource indicators, and a standalone `/connect` route; native Anthropic `/v1/messages` routed through the Rust axum gateway alongside a Responses API WebSockets port and SigV4 in `litellm-core`; SAML 2.0 SSO for the admin UI; roughly twenty dashboard routes completing the shadcn and shared DataTable migration; and a budget-reset correctness pass covering reset time of day, word-form durations, and NULL `budget_reset_at` repair. Note that user budgets are no longer enforced on team keys, reverting the `v1.94.0` change.
+An internal user's `object_permission` becomes a real MCP entitlement level that intersects the key, team, agent, and org scopes and is enforced at both `tools/list` and `tools/call`; a new `post_mcp_call` guardrail mode that finally lets a guardrail mask or reject the contents of an MCP tool result; config changes propagating to every pod through a coordination-Redis invalidation event instead of the 30s poll; OpenAI's GPT-5.6 price cut of 20% on terra and 80% on luna, mirrored onto Bedrock Mantle with new flex long-context rates; a complexity classifier that now sees prior and assistant turns and records its tier decision in spend logs; a generic `/management/v1` list contract with `GET /management/v1/budgets` on top of it; and operational hardening for large deployments covering DB statement and lock timeouts, `REPLICA IDENTITY FULL`, and an unreachable Redis that no longer blocks every request. Note that mock testing request params are now gated behind a config flag that is off by default.
 
 ---
 
 ## Recent Releases
 
 | Version                             | Date         | Highlights                                                 |
 | ----------------------------------- | ------------ | ---------------------------------------------------------- |
+| [v1.96.0](/release_notes/v1.96.0/v1-96-0)   | Aug 9, 2026  | MCP entitlements, Redis config sync, auto-router context, GPT-5.6 price cut |
 | [v1.95.0](/release_notes/v1.95.0/v1-95-0)   | Aug 1, 2026  | Claude Opus 5, MCP gateway DCR, Rust `/v1/messages`, SAML 2.0 SSO |
 | [v1.94.0](/release_notes/v1.94.0/v1-94-0)   | Jul 28, 2026 | Router plugins & Auto-Router v2, MCP client-held credentials, shared DataTable UI |
 | [v1.93.0](/release_notes/v1.93.0/v1-93-0)   | Jul 18, 2026 | GPT-5.6, client-forwarded MCP credentials, Meta Model API provider |

diff --git a/release_notes/v1.96.0rc1/index.md → release_notes/v1.96.0/index.md b/release_notes/v1.96.0rc1/index.md → release_notes/v1.96.0/index.md
@@ -1,7 +1,7 @@
 ---
-title: "v1.96.0rc1 - MCP Entitlements, Redis Config Sync & Auto-Router Context"
-slug: "v1-96-0-rc-1"
-date: 2026-08-03T12:53:08
+title: "v1.96.0 - MCP Entitlements, Redis Config Sync & Auto-Router Context"
+slug: "v1-96-0"
+date: 2026-08-09T00:00:00
 authors:
   - name: Krrish Dholakia
     title: CEO, LiteLLM
@@ -30,22 +30,28 @@ import TabItem from '@theme/TabItem';
 docker run \
 -e STORE_MODEL_IN_DB=True \
 -p 4000:4000 \
-docker.litellm.ai/berriai/litellm:1.96.0-rc.1
+docker.litellm.ai/berriai/litellm:1.96.0
 ```
 
 </TabItem>
 <TabItem value="pip" label="Pip">
 
 ```bash
-pip install litellm==1.96.0rc1
+pip install litellm==1.96.0
 ```
 
 </TabItem>
 </Tabs>
 
-## Key Highlights
+:::danger Breaking Changes
+
+**Mock testing request params are gated behind one config flag.** Six `mock_*` request params previously had three different behaviors, and a dropped param returned a normal success, so a fallback drill could pass without ever running. All six now reject with a 400 naming the params and the key unless an admin sets `general_settings.dangerously_allow_mock_testing_request_params: true` in `config.yaml`; the flag cannot be changed from the Admin UI or the API. See [PR #35423](https://github.com/BerriAI/litellm/pull/35423).
 
-`v1.96.0rc1` is the current release candidate for 1.96.0.
+**Keyless gateway OAuth now admits session bearers at any MCP scope.** Session-bearer admission and RFC 9728 `WWW-Authenticate` challenges fire on per-server MCP URL paths, not only the aggregate `/mcp/` scope, so a per-server path that previously fell through to a plain rejection now issues a challenge. See [PR #34856](https://github.com/BerriAI/litellm/pull/34856).
+
+:::
+
+## Key Highlights
 
 - **MCP entitlements reach the person, not just the key** - an internal user's `object_permission` now acts as an MCP entitlement level that intersects the key, team, agent, and org scopes, is read at both `tools/list` and `tools/call` time, is persisted by `/user/new` and `/user/update`, returned by `/v2/user/info`, and editable from the internal user page.
 - **Guardrails can finally see MCP tool results** - a new `post_mcp_call` mode routes tool result text through the unified `apply_guardrail` seam, so a guardrail can mask values inside a result or reject it outright; previously a tool returning sensitive data bypassed every guardrail.
@@ -55,12 +61,6 @@ pip install litellm==1.96.0rc1
 - **Budgets become a first-class management surface** - a generic `/management/v1` list contract lands with `GET /management/v1/budgets` on top of it, and the budgets page gains sorting, filtering, and search.
 - **Operational hardening for large deployments** - opt-in `database_statement_timeout` and `database_lock_timeout`, opt-in `REPLICA IDENTITY FULL` re-asserted after every migration, pod-hardening and migration-Job knobs on the componentized Helm chart, and an unreachable Redis that no longer blocks every request.
 
-### Breaking changes
-
-- **Mock testing params are gated behind one config flag** - six mock testing request params previously had three different behaviors, and a dropped param returned a normal success, so a fallback drill could pass without ever running. All six are now gated by a single config flag that is unset by default and rejects with a 400 naming the params and the key - [PR #35423](https://github.com/BerriAI/litellm/pull/35423)
-- **User budgets are no longer enforced on team keys** - reverts [PR #32005](https://github.com/BerriAI/litellm/pull/32005), which caused team keys to reject on the key owner's personal budget; team keys use team budgets only and the `skip_user_budget_on_team_key` opt-out flag is removed - [PR #35271](https://github.com/BerriAI/litellm/pull/35271)
-- **Keyless gateway OAuth admits session bearers at any MCP scope** - session-bearer admission and RFC 9728 `WWW-Authenticate` challenges now fire on per-server MCP URL paths, not only the aggregate `/mcp/` scope - [PR #34856](https://github.com/BerriAI/litellm/pull/34856)
-
 ## New Models / Updated Models
 
 #### New Model Support (2 new models)
@@ -297,4 +297,4 @@ This window added 18 test-only PRs, 10 of them against the live e2e suite. New a
 
 ## Full Changelog
 
-https://github.com/BerriAI/litellm/compare/v1.95.0-rc.1...v1.96.0-rc.1
+https://github.com/BerriAI/litellm/compare/v1.95.0...v1.96.0