docs(release-notes): add v1.82.6.rc.1 release notes

[joereyna]

Summary

• Adds full release notes for v1.82.6.rc.1 covering 128 PRs from v1.82.3-stable
• Follows the release notes guide structure (Deploy → Key Highlights → Providers/Models → LLM API → Management → AI Integrations → MCP Gateway → Spend Tracking → Performance → Security → Documentation Updates → New Contributors → Full Changelog)

Key Highlights Covered

• gpt-5.4-mini & gpt-5.4-nano — day-0 support (also Azure variants)
• Volcengine Doubao Seed 2.0 — pro/lite/mini/code-preview with tiered pricing
• Multi-proxy control plane (feat: add control plane for multi-proxy worker management #24217) — manage multiple LiteLLM proxies from one admin UI
• Privilege escalation fix ([Fix] Privilege Escalation on /key/block, /key/unblock, and /key/update max_budget #23781) — users could no longer escalate their own role
• Anthropic reasoning summary opt-out (feat(anthropic): add opt-out flag for default reasoning summary #22904)
• Prompt management for Responses API ([feat]Add prompt management support for responses api #23999)
• Per-model-group deployment affinity (feat(router): add per-model-group deployment affinity #24110)
• MCP Python SDK upgrade to v1.26.0 (fix: upgrade mcp to 1.26.0 and bump version to 1.82.5 #24179)
• Global secret redaction fix for all log paths (fix: global secret redaction via root logger + key-name-based pattern matching #24305)
• Akto guardrail integration (Add Akto Guardrails to LiteLLM #23250)
• MCP JWT Signer zero-trust guardrail (feat(guardrails): MCPJWTSigner - built-in guardrail for zero trust MCP auth #23897)
• Interactive setup wizard litellm --setup (feat(setup): interactive setup wizard + install.sh #23644)

Test plan

• Verify markdown renders correctly on docs site
• Confirm all PR links are valid
• Check new contributor GitHub handles are accurate

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Mar 24, 2026 5:24pm

[codspeed-hq]

Merging this PR will not alter performance

✅ 16 untouched benchmarks

---

_{Comparing joereyna:release-notes/v1.82.6 (87c4c27) with main (0af114f)}

[greptile-apps]

Greptile Summary

This PR adds release notes for v1.82.6.rc.1, covering 128 merged PRs since v1.82.3-stable. The document follows the established release notes structure and accurately documents key highlights including day-0 support for gpt-5.4-mini/gpt-5.4-nano, the multi-proxy control plane, a privilege escalation security fix, and the Volcengine Doubao Seed 2.0 model family.

Issues found:

• PR #23999 (Responses API prompt management) is listed twice — once under LLM API Endpoints → Features (line 156) and again under AI Integrations → Prompt Management (line 232) with slightly different descriptions. Consider consolidating to one canonical entry and adding a cross-reference in the other section.
• The Diff Summary counts appear inconsistent: Management Endpoints / UI: 17 vs ~20 countable items, and Logging / Guardrail / Prompt Management Integrations: 9 vs 10 entries when the Prompt Management sub-section is included.

Confidence Score: 4/5

• Safe to merge — documentation-only change with no functional code modifications.
• The PR adds a single Markdown file with no executable code. The content is well-structured and mostly accurate; the two remaining issues (duplicate PR entry and minor count discrepancies in the summary table) are cosmetic and do not affect users or the build.
• docs/my-website/release_notes/v1.82.6/index.md — verify PR [feat]Add prompt management support for responses api #23999 duplication and Diff Summary counts.

Important Files Changed

Filename	Overview
docs/my-website/release_notes/v1.82.6/index.md	New release notes file for v1.82.6.rc.1 covering 128 PRs; well-structured and mostly consistent, with two minor issues: PR #23999 duplicated across LLM API Endpoints and AI Integrations sections with slightly different descriptions, and the Diff Summary counts for Management Endpoints/UI (17 vs ~20 items) and Logging/Guardrail/Prompt Management (9 vs 10 items) appear inaccurate.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[v1.82.6 Release Notes] --> B[Deploy]
    A --> C[Key Highlights]
    A --> D[New Models / Updated Models]
    A --> E[LLM API Endpoints]
    A --> F[Management Endpoints / UI]
    A --> G[AI Integrations]
    A --> H[MCP Gateway]
    A --> I[Spend Tracking & Rate Limiting]
    A --> J[Performance / Reliability]
    A --> K[Security]
    A --> L[New Contributors]
    A --> M[Diff Summary]

    D --> D1[12 New Models\ngpt-5.4-mini/nano, xAI grok-4.20, Volcengine]
    D --> D2[Updated Models\nOpenAI, xAI, Azure AI, Vertex AI]
    D --> D3[Features & Bugs]

    E --> E1[Features\nVideo Gen, Responses API, Azure]
    E --> E2[Bugs\nStreaming logging]

    G --> G1[Logging\nLangfuse, LangSmith]
    G --> G2[Guardrails\nAkto, MCP JWT Signer]
    G --> G3[Prompt Management\nPR #23999 ⚠️ also in E1]

    K --> K1[Privilege Escalation Fix\nPR #23781]
    K --> K2[Global Secret Redaction\nPR #24305]

Loading

_{Reviews (4): Last reviewed commit: "docs(release-notes): fix remaining grept..." | Re-trigger Greptile}

[greptile-apps]

Incorrect context window for azure/gpt-5.4-mini

The azure/gpt-5.4-mini row lists a context window of 1.05M, but that value belongs to gpt-5.4-nano. The Key Highlights section on line 44 explicitly states gpt-5.4-mini has a 272K context, and the OpenAI row on line 60 also correctly shows 272K. This looks like the nano row's value was accidentally copied.

Suggested change

	| Azure OpenAI | `azure/gpt-5.4-mini` | 1.05M | $0.75 | $4.50 | chat, vision, tools, reasoning |
	| Azure OpenAI | `azure/gpt-5.4-mini` | 272K | $0.75 | $4.50 | chat, vision, tools, reasoning |

[greptile-apps]

MCP SDK version/PR discrepancy with PR description

The file credits the mcp SDK upgrade to v1.26.0 via PR #24179, but the PR description lists "MCP Python SDK upgrade to v1.9.0 (#24102)" as a key highlight. PR #24102 does not appear anywhere in the release notes file.

This could mean either:

• PR [Feature]: add an option to disable prefixing each tool name with its MCP server name #24102 (v1.9.0 upgrade) was merged separately from PR fix: upgrade mcp to 1.26.0 and bump version to 1.82.5 #24179 (v1.26.0 upgrade), and PR [Feature]: add an option to disable prefixing each tool name with its MCP server name #24102 is missing from the notes.
• The PR description was generated with incorrect metadata and v1.26.0 / PR fix: upgrade mcp to 1.26.0 and bump version to 1.82.5 #24179 is the correct reference.

Please verify whether PR #24102 should also appear in the MCP Gateway section (or if v1.26.0 superseded the v1.9.0 bump in a single PR).

[greptile-apps]

xAI model listed under OpenAI section

xai/grok-beta is an xAI model, not an OpenAI model. Listing it under the [OpenAI] heading is misleading for readers scanning the "Updated Models" section. Consider either:

• Moving it to an [xAI] subsection of "Updated Models", or
• Grouping it with the three new xAI models already listed in the "New Models" table

Suggested change

	- Add `supports_minimal_reasoning_effort` to `xai/grok-beta`
	- Add `supports_minimal_reasoning_effort` to entire `gpt-5.x` model series (gpt-5.1 through gpt-5.4, including codex, pro, nano, and mini variants) and `azure/gpt-5.1-2025-11-13`
	- **[xAI](../../docs/providers/xai)**
	- Add `supports_minimal_reasoning_effort` to `xai/grok-beta`

_{Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!}

[greptile-apps]

PR number mismatch with PR description key highlight

The PR description calls out "Global secret redaction fix for streaming responses (#24128)" as a key highlight, but this file references PR #24305 for the global secret redaction fix, and PR #24128 does not appear anywhere in the release notes.

This could mean:

• PR fix: strip $schema from tool input_schema in beta /v1/messages endpoint #24128 is the streaming-specific fix that is separate from the root-logger fix in fix: global secret redaction via root logger + key-name-based pattern matching #24305, and it is missing from the notes.
• The PR description metadata was generated from stale context and fix: global secret redaction via root logger + key-name-based pattern matching #24305 is the canonical fix.

Please confirm whether PR #24128 should also be listed here (or in the General Bugs section), or whether the PR description should be updated to reference #24305.

[greptile-apps]

PR #23926 appears in both Features and Bugs sections

PR #23926 is listed twice with slightly different descriptions:

• Line 90 (Features → OpenAI): "Auto-route gpt-5.4+ calls using both tools and reasoning to the Responses API on Azure"
• Line 121 (Bugs → Azure AI): "Auto-route Azure gpt-5.4+ tools+reasoning calls to Responses API"

The same PR appearing in both a Features section and a Bugs section creates inconsistency — it was either a new feature or a bug fix. Please consolidate into the most appropriate section and remove the duplicate.

[greptile-apps]

Diff Summary count for LLM API Endpoints appears inaccurate

The Diff Summary states "LLM API Endpoints: 6", but counting the entries in that section yields only 4 items:

• 3 features (Video Generation API, Responses API prompt management, Azure AZURE_DEFAULT_API_VERSION)
• 1 bug fix (logging for incomplete streaming)

Please verify whether items from other sections were intended to be counted here, or update the count to 4.

Suggested change

	* LLM API Endpoints: 6
	* LLM API Endpoints: 4

[greptile-apps]

azure_ai/DeepSeek-V3-2-speciale — possible typo in model name

The model name azure_ai/DeepSeek-V3-2-speciale uses the Italian/Spanish word "speciale" rather than the English "special". If the actual Azure AI model slug is DeepSeek-V3-2-special or a different variant, this will direct users to a non-existent model name. Please verify this is the exact model identifier published by Azure AI.

[greptile-apps]

SDK upgrade miscategorized as a bug fix

Upgrade mcp SDK to 1.26.0 is listed under the #### Bugs header in the MCP Gateway section. Dependency version bumps are maintenance tasks or features, not bug fixes. Listing it here may mislead users scanning the notes for actual bug resolutions.

Consider moving it to a dedicated #### Maintenance or #### Upgrades subsection, or (if the upgrade itself fixes a bug) briefly noting the specific defect it resolves so the categorization is justified.

Suggested change

	- Upgrade `mcp` SDK to 1.26.0 - [PR #24179](https://github.com/BerriAI/litellm/pull/24179)
	- Upgrade `mcp` SDK to 1.26.0 - [PR #24179](https://github.com/BerriAI/litellm/pull/24179)

Move this entry under a #### Features or #### Maintenance header instead of #### Bugs.

_{Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!}

[greptile-apps]

PR #23167 listed twice with slightly different descriptions

PR #23167 "Export audit logs to external callback systems" appears in two separate sections:

• Line 176 (Management Endpoints / UI → Audit Logs): "Export audit logs to external callback systems (S3, custom callbacks)"
• Line 218 (AI Integrations → Logging → Audit Log Export): "Export audit logs to external callback systems (S3, custom destinations)"

The differing wording ("custom callbacks" vs. "custom destinations") for the same PR creates inconsistency for readers. If the feature genuinely spans both sections, consider keeping the primary entry in one section and adding a brief cross-reference in the other, or standardising the description.

_{Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!}

[greptile-apps]

PR #23999 listed twice with different descriptions

PR #23999 is described twice across two separate sections with slightly different wording:

• Line 156 (LLM API Endpoints → Features → Responses API): "Prompt management support for Responses API — use prompt templates and versioning with /v1/responses"
• Line 232 (AI Integrations → Prompt Management → Responses API): "Prompt management (templates, versioning) now supported for /v1/responses"

This creates the same redundancy pattern already flagged for PR #23167. If the intent is to surface this feature in multiple sections for discoverability, consider keeping the full entry in one section (e.g., LLM API Endpoints where the endpoint change lives) and replacing the second occurrence with a brief cross-reference like "See LLM API Endpoints → Responses API above."

_{Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!}

[greptile-apps]

Diff Summary counts appear inconsistent

Two counts in the Diff Summary section look off:

1. Management Endpoints / UI: 17 — Counting the bullets directly yields 6 feature entries + 14 bug entries = 20 items (even counting the Teams table bullet with two PRs as a single entry). It's unclear how 17 is derived.

2. Logging / Guardrail / Prompt Management Integrations: 9 — The section heading includes "Prompt Management", but counting its entries gives: 4 logging + 5 guardrail + 1 prompt management = 10. If prompt management is excluded from the count (because PR [feat]Add prompt management support for responses api #23999 is already counted under LLM API Endpoints), the label should be updated to omit "Prompt Management" to avoid confusion.

Please verify and reconcile these totals so the summary table accurately reflects the section contents.