A lot of Bedrocks newest models are only launching on Bedrock Mantle and not Bedrock runtime. GPT 5.5 and Gemma 4 for instance.
I can’t see anywhere in the docs on how to use Mantle with ChatBrdrockConverse or some equivalent. Is Bedrock Mantle not supported?
You’re right that there’s a gap here. Per the LangChain docs, ChatBedrockConverse targets the Bedrock Runtime Converse API, not Bedrock Mantle. Mantle isn’t documented or supported as a first-class integration in langchain-aws today.
Bedrock Mantle is a separate OpenAI-compatible inference layer (AWS announcement). Models like GPT 5.5 and Gemma 4 that launch only on Mantle won’t work through ChatBedrockConverse.
Workaround today: use ChatOpenAI with Mantle’s base URL, per the OpenAI-compatible endpoints docs:
from langchain_openai import ChatOpenAI
llm = ChatOpenAI(
model="openai.gpt-oss-120b", # use the Mantle model ID
base_url="https://bedrock-mantle.us-east-1.api.aws/v1",
api_key="...", # Bedrock API key, or configure AWS SigV4 auth
)
This works for basic chat completions. Caveats:
ChatOpenAI only implements the standard OpenAI spec, Mantle-specific fields (e.g. reasoning) won’t be extracted
- No built-in support for Mantle’s Responses API, Projects API, or other Mantle-only features
- Auth may need extra setup for IAM/SigV4 vs. a Bedrock API key
Bottom line: Mantle isn’t officially supported yet. The OpenAI-compatible route is the practical option for now. A dedicated ChatBedrockMantle (or similar) in langchain-aws would be the right long-term fix, worth opening a feature request on langchain-ai/langchain-aws if you need first-class support.
@keenborder786 Models like GPT-5.5 aren’t supported by ChatOpenAI because AWS Bedrock Mantle doesn’t provide the Chat Completions API for them: GPT-5.5 - Amazon Bedrock .
ChatOpenAI works only with models that support the Chat Completions API, such as gpt-oss-120b and gpt-oss-20b.
You are right that GPT‑5.5 is a gap; on Mantle it’s Responses API only (.../openai/v1), so ChatOpenAI with default Chat Completions routing won’t reach it.
However, Gemma 4 is still different, though: Mantle exposes it on both Chat Completions and Responses (AWS blog). So ChatOpenAI(model="google.gemma-4-31b", base_url="https://bedrock-mantle.{region}.api.aws/openai/v1", api_key="<bedrock-key>") should work today via Chat Completions.
However, I would suggest you try the Responses API. You can experiment with:
ChatOpenAI(
model="openai.gpt-5.5",
use_responses_api=True,
base_url="https://bedrock-mantle.{region}.api.aws/openai/v1",
api_key="<bedrock-api-key>",
)
This is undocumented for Mantle and only covers Bearer/API‑key auth (IAM/SigV4 won’t work through ChatOpenAI). There’s no first‑class Mantle integration in langchain-aws yet, so a feature request on langchain-ai/langchain-aws is STILL the right long‑term path.