feat: add GPT-5.6 Terra via bedrock-mantle Responses API

agent/Dockerfile

@@ -13,7 +13,7 @@ RUN uv pip install --no-cache -r requirements.txt
 RUN useradd -m -u 1000 bedrock_agentcore
 USER bedrock_agentcore
 
-COPY basic_agent.py cost_logger.py factory.py composition.py mcp_clients.py mantle_client.py model_profiles.py resilience.py session.py streaming.py ./
+COPY basic_agent.py cost_logger.py errors.py factory.py composition.py mcp_clients.py mantle_client.py model_profiles.py resilience.py session.py streaming.py ./
 COPY prompts/ prompts/
 COPY modes/ modes/
 COPY tools/ tools/

agent/mantle_client.py

@@ -40,12 +40,28 @@ async def aclose(self):
 
 
 def mantle_model(model_id: str, region: str = "us-east-1"):
-    """Create a Strands OpenAIModel pointing at bedrock-mantle with SigV4 auth.
+    """Create a Strands model pointing at bedrock-mantle with AWS-native auth.
+
+    Models that only support the Responses API (see
+    ``model_profiles.MANTLE_RESPONSES_MODELS``, e.g. GPT-5.6 Terra) use
+    Strands' native ``OpenAIResponsesModel`` with ``bedrock_mantle_config``
+    (bearer token minted per request). All other models use the legacy
+    Chat Completions path with a SigV4-signed httpx transport.
 
     Usage:
         from mantle_client import mantle_model
         model = mantle_model("openai.gpt-5.4", region="us-east-1")
     """
+    from model_profiles import MANTLE_RESPONSES_MODELS
+
+    if model_id in MANTLE_RESPONSES_MODELS:
+        from strands.models.openai_responses import OpenAIResponsesModel
+
+        return OpenAIResponsesModel(
+            bedrock_mantle_config={"region": region},
+            model_id=model_id,
+        )
+
     from strands.models.openai import OpenAIModel
 
     class _MantleOpenAIModel(OpenAIModel):

agent/model_profiles.py

@@ -99,10 +99,18 @@ def to_bedrock_kwargs(self) -> dict:
 # Models served via bedrock-mantle (OpenAI-compatible endpoint, not Converse API).
 # model_id → list of supported regions (first entry is the fallback).
 MANTLE_MODELS: dict[str, list[str]] = {
+    "openai.gpt-5.6-terra": ["us-east-1", "us-east-2", "us-west-2"],
     "openai.gpt-5.5": ["us-east-1", "us-east-2"],
     "openai.gpt-5.4": ["us-east-1", "us-east-2", "us-west-2"],
 }
 
+# Mantle models that only support the Responses API (no Chat Completions).
+# GPT-5.6 model cards: Responses ✅ / Chat Completions ❌ — calling
+# /chat/completions returns 400 Bad Request.
+MANTLE_RESPONSES_MODELS: set[str] = {
+    "openai.gpt-5.6-terra",
+}
+
 
 def resolve_mantle_region(model_id: str, deploy_region: str | None = None) -> str:
     """Resolve the best mantle region for a model.
@@ -136,6 +144,7 @@ def resolve_mantle_region(model_id: str, deploy_region: str | None = None) -> st
     # Amazon Nova
     "us.amazon.nova-2-lite-v1:0": NOVA_2_DEFAULT,
     # OpenAI GPT (bedrock-mantle)
+    "openai.gpt-5.6-terra": GPT_DEFAULT,
     "openai.gpt-5.5": GPT_DEFAULT,
     "openai.gpt-5.4": GPT_DEFAULT,
 }

agent/requirements.txt

@@ -1,6 +1,6 @@
 # Dependencies for Strands Agent. AI model access requires appropriate
 # IAM permissions and Bedrock model access grants.
-strands-agents[otel,openai]>=1.30.0
+strands-agents[otel,openai]>=1.47.0
 mcp>=1.23.1
 bedrock-agentcore[strands-agents]>=1.0.6
 aws-opentelemetry-distro>=0.10.0

infra/config.example.yaml

@@ -42,6 +42,7 @@ model:
     - "global.anthropic.claude-opus-4-6-v1"
     - "global.anthropic.claude-haiku-4-5-20251001-v1:0"
     - "us.amazon.nova-2-lite-v1:0"
+    - "openai.gpt-5.6-terra"                         # bedrock-mantle (us-east-1, us-east-2, us-west-2)
     - "openai.gpt-5.5"                               # bedrock-mantle (us-east-1, us-east-2)
     - "openai.gpt-5.4"                               # bedrock-mantle (us-east-1, us-east-2, us-west-2)
 

infra/lib/agent-stack.ts

@@ -65,10 +65,12 @@ export class AgentStack extends cdk.Stack {
         resources: ["*"],
       })
     );
-    // bedrock-mantle (OpenAI-compatible endpoint) for models like GPT-5.4/5.5
+    // bedrock-mantle (OpenAI-compatible endpoint) for models like GPT-5.4/5.5/5.6.
+    // CreateInference: legacy SigV4 path (Chat Completions).
+    // CallWithBearerToken: Responses API path (bearer token, e.g. GPT-5.6 Terra).
     role.addToPolicy(
       new iam.PolicyStatement({
-        actions: ["bedrock-mantle:CreateInference"],
+        actions: ["bedrock-mantle:CreateInference", "bedrock-mantle:CallWithBearerToken"],
         resources: ["*"],
       })
     );

infra/lib/model-metadata.ts

@@ -50,6 +50,10 @@ export const MODEL_METADATA: Record<string, ModelMetadata> = {
     composable: false,
   },
   // --- OpenAI GPT ---
+  "openai.gpt-5.6-terra": {
+    displayName: "GPT-5.6 Terra",
+    description: "Balanced performance competitive with GPT-5.5 at half the cost",
+  },
   "openai.gpt-5.5": {
     displayName: "GPT-5.5",
     description: "OpenAI's most capable model, advanced coding and agentic tasks",