docs(bedrock): gpt-5.6 and newer default to the native chat completions route, chat_completions/ opts the rest in

[mateo-berri]

Companion docs for BerriAI/litellm#44307, which sends Bedrock GPT-5.6 and newer chat completions (the GPT-5.6, GPT-6, and GPT-6.1 families today) to bedrock-runtime's native /openai/v1/chat/completions endpoint by default and adds a bedrock/chat_completions/ prefix that opts any other model AWS serves there (GPT-OSS and Grok 4.6) into the same route. The Bedrock provider page gets a "Native Chat Completions route" section naming which models go native by default and which need the prefix, the cost map condition behind the default (a row listing /v1/chat/completions under supported_endpoints), the per-request cases that still use Converse (guardrailConfig and the other Converse body fields, operator request metadata, application inference profiles, function tools with reasoning on, thinking on /chat/completions, and an explicit bedrock/converse/ prefix) with the reason for each, and what a caller will notice in the response. The GPT-OSS section keeps its bedrock/converse/openai.gpt-oss-* examples, since those models stay on Converse unless prefixed, and points at the prefix for the native route

Merge after BerriAI/litellm#44307 lands so the page does not describe an unreleased route

Caveats

Low: the structure check (not required on main) is red at 2b6b794 on docs/harness/quickstart.md and docs/harness/deepagents.md, files this PR does not touch, and main fixed it in #1951 (f1fda66), newer than this branch's merge base a134432, so it is left as is

---

Note

Low Risk
Documentation-only changes with no runtime impact; accuracy depends on the companion implementation PR shipping first.

Overview
Documents companion behavior for the upcoming Bedrock native /openai/v1/chat/completions routing (merge after the main LiteLLM PR lands).

The Bedrock provider page is updated to call out bedrock/chat_completions/ and note that bedrock/ defaults to native Chat Completions for GPT-5.6 and newer. A new Native Chat Completions route section explains which model families use native vs Converse by default, how the model cost map’s supported_endpoints drives the default, per-request Converse fallbacks (guardrails, Converse-only body fields, tools + reasoning, JSON modes, etc.), and response/parameter differences on the native path.

The converse/invoke section adds a short GPT-5.6+ exception, with SDK/proxy examples for opting GPT-OSS/Grok in via bedrock/chat_completions/ and pinning newer GPT models to bedrock/converse/. The OpenAI GPT OSS table now links to the native route prefix while keeping existing Converse examples.

^{Reviewed by Cursor Bugbot for commit 2b6b794. Bugbot is set up for automated code reviews on this repo. Configure here.}

[vercel]

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

Project	Deployment	Actions	Updated
litellm	Ready	Preview	Sep 30, 2026 11:58pm UTC

[mateo-berri]

bugbot run

[cursor]

Stale Bugbot comment from a previous run.

[mateo-berri]

bugbot run

[cursor]

Stale Bugbot comment from a previous run.

[mateo-berri]

bugbot run

[cursor]

Stale Bugbot comment from a previous run.

[mateo-berri]

bugbot run

[cursor]

Bugbot is paused — on-demand spend limit reached

Bugbot uses usage-based billing for this team and has hit its on-demand spend limit.

A team admin can raise the spend limit in the Cursor dashboard, or wait for the next billing cycle to continue.

[mateo-berri]

bugbot run

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 2b6b794. Configure here.}