ci: switch to another model for aws bedrock real provider tests in ci (#2389)

**Description**
The AWS Bedrock e2e and data-plane tests were failing `because
us.meta.llama3-2-1b-instruct-v1:0` reached end-of-life on Bedrock (404
ResourceNotFoundException). Switched the AWS Bedrock test model to the
cheapest active option, Amazon Nova Micro (us.amazon.nova-micro-v1:0).

Also centralized all real-provider model IDs so future model churn is a
one-line fix.

More details are on #2388

Signed-off-by: yxia216 <yxia216@bloomberg.net>

Author: hustxiayang

examples/basic/aws-irsa.yaml

@@ -105,7 +105,7 @@ spec:
         - headers:
             - type: Exact
               name: x-ai-eg-model
-              value: us.meta.llama3-2-1b-instruct-v1:0
+              value: us.amazon.nova-micro-v1:0
       backendRefs:
         - name: envoy-ai-gateway-basic-aws
 ---

examples/basic/aws-pod-identity.yaml

@@ -100,7 +100,7 @@ spec:
         - headers:
             - type: Exact
               name: x-ai-eg-model
-              value: us.meta.llama3-2-1b-instruct-v1:0
+              value: us.amazon.nova-micro-v1:0
       backendRefs:
         - name: envoy-ai-gateway-basic-aws
 ---

examples/basic/aws.yaml

@@ -18,7 +18,7 @@ spec:
         - headers:
             - type: Exact
               name: x-ai-eg-model
-              value: us.meta.llama3-2-1b-instruct-v1:0
+              value: us.amazon.nova-micro-v1:0
       backendRefs:
         - name: envoy-ai-gateway-basic-aws
 ---

examples/provider_fallback/base.yaml

@@ -42,7 +42,7 @@ spec:
         - headers:
             - type: Exact
               name: x-ai-eg-model
-              value: us.meta.llama3-2-1b-instruct-v1:0
+              value: us.amazon.nova-micro-v1:0
       backendRefs:
         - name: provider-fallback-always-failing-upstream # This is the primary backend and trying to speak TLS, which always fails.
           priority: 0

internal/testing/real_provider_models.go

@@ -0,0 +1,62 @@
+// Copyright Envoy AI Gateway Authors
+// SPDX-License-Identifier: Apache-2.0
+// The full text of the Apache license is available in the LICENSE file at
+// the root of the repo.
+
+package internaltesting
+
+// Model IDs used by tests that exercise real provider backends.
+//
+// These are centralized here so that when a provider retires a model (which
+// surfaces as a 4xx/deprecation error from the real API), the swap is a
+// single-line change instead of hunting down duplicated string literals across
+// the test suite.
+//
+// Note: AWS Bedrock example manifests (e.g. examples/basic/aws.yaml,
+// examples/provider_fallback/base.yaml, tests/data-plane/envoy.yaml) route by
+// the exact model name and must be kept in sync manually, as they cannot
+// reference these Go constants.
+const (
+	// Chat completion models (one per real provider).
+
+	// OpenAIModelName is the OpenAI chat completion model.
+	OpenAIModelName = "gpt-4o-mini"
+	// AWSBedrockModelName is the AWS Bedrock chat completion model.
+	AWSBedrockModelName = "us.amazon.nova-micro-v1:0"
+	// AzureOpenAIModelName is the Azure OpenAI chat completion model.
+	AzureOpenAIModelName = "o1"
+	// GeminiModelName is the Gemini chat completion model.
+	GeminiModelName = "gemini-3.1-flash-lite"
+	// GroqModelName is the Groq chat completion model.
+	GroqModelName = "llama-3.1-8b-instant"
+	// GrokModelName is the Grok (xAI) chat completion model.
+	GrokModelName = "grok-3"
+	// SambaNovaModelName is the SambaNova chat completion model.
+	SambaNovaModelName = "Meta-Llama-3.1-8B-Instruct"
+	// DeepInfraModelName is the DeepInfra chat completion model.
+	DeepInfraModelName = "meta-llama/Meta-Llama-3-8B-Instruct"
+
+	// Embeddings models (one per real provider that supports embeddings).
+
+	// OpenAIEmbeddingsModelName is the OpenAI embeddings model.
+	OpenAIEmbeddingsModelName = "text-embedding-3-small"
+	// AWSBedrockEmbeddingsModelName is the AWS Bedrock embeddings model.
+	AWSBedrockEmbeddingsModelName = "amazon.titan-embed-text-v2:0"
+	// GeminiEmbeddingsModelName is the Gemini embeddings model.
+	GeminiEmbeddingsModelName = "gemini-embedding-001"
+	// SambaNovaEmbeddingsModelName is the SambaNova embeddings model.
+	SambaNovaEmbeddingsModelName = "E5-Mistral-7B-Instruct"
+	// DeepInfraEmbeddingsModelName is the DeepInfra embeddings model.
+	DeepInfraEmbeddingsModelName = "BAAI/bge-base-en-v1.5"
+
+	// Anthropic Messages API models.
+
+	// AnthropicModelName is the direct Anthropic API model.
+	AnthropicModelName = "claude-sonnet-4-5"
+	// AWSBedrockAnthropicGlobalModelName is the Claude model served by AWS
+	// Bedrock via the global cross-region inference profile.
+	AWSBedrockAnthropicGlobalModelName = "global.anthropic.claude-sonnet-4-5-20250929-v1:0"
+	// AWSBedrockAnthropicUSModelName is the Claude model served by AWS Bedrock
+	// via the US cross-region inference profile.
+	AWSBedrockAnthropicUSModelName = "us.anthropic.claude-sonnet-4-5-20250929-v1:0"
+)

tests/data-plane/envoy.yaml

@@ -58,7 +58,7 @@ static_resources:
                             headers:
                               - name: x-ai-eg-model
                                 string_match:
-                                  exact: us.meta.llama3-2-1b-instruct-v1:0
+                                  exact: us.amazon.nova-micro-v1:0
                           route:
                             auto_host_rewrite: true
                             cluster: aws_bedrock

tests/data-plane/real_providers_test.go

@@ -103,14 +103,14 @@ func TestWithRealProviders(t *testing.T) {
 	t.Run("health-checking", func(t *testing.T) {
 		t.Run("chat/completions", func(t *testing.T) {
 			for _, tc := range []realProvidersTestCase{
-				{name: "openai", modelName: "gpt-4o-mini", required: internaltesting.RequiredCredentialOpenAI},
-				{name: "aws-bedrock", modelName: "us.meta.llama3-2-1b-instruct-v1:0", required: internaltesting.RequiredCredentialAWS},
-				{name: "azure-openai", modelName: "o1", required: internaltesting.RequiredCredentialAzure},
-				{name: "gemini", modelName: "gemini-3.1-flash-lite", required: internaltesting.RequiredCredentialGemini},
-				{name: "groq", modelName: "llama-3.1-8b-instant", required: internaltesting.RequiredCredentialGroq},
-				{name: "grok", modelName: "grok-3", required: internaltesting.RequiredCredentialGrok},
-				{name: "sambanova", modelName: "Meta-Llama-3.1-8B-Instruct", required: internaltesting.RequiredCredentialSambaNova},
-				{name: "deepinfra", modelName: "meta-llama/Meta-Llama-3-8B-Instruct", required: internaltesting.RequiredCredentialDeepInfra},
+				{name: "openai", modelName: internaltesting.OpenAIModelName, required: internaltesting.RequiredCredentialOpenAI},
+				{name: "aws-bedrock", modelName: internaltesting.AWSBedrockModelName, required: internaltesting.RequiredCredentialAWS},
+				{name: "azure-openai", modelName: internaltesting.AzureOpenAIModelName, required: internaltesting.RequiredCredentialAzure},
+				{name: "gemini", modelName: internaltesting.GeminiModelName, required: internaltesting.RequiredCredentialGemini},
+				{name: "groq", modelName: internaltesting.GroqModelName, required: internaltesting.RequiredCredentialGroq},
+				{name: "grok", modelName: internaltesting.GrokModelName, required: internaltesting.RequiredCredentialGrok},
+				{name: "sambanova", modelName: internaltesting.SambaNovaModelName, required: internaltesting.RequiredCredentialSambaNova},
+				{name: "deepinfra", modelName: internaltesting.DeepInfraModelName, required: internaltesting.RequiredCredentialDeepInfra},
 			} {
 				t.Run(tc.name, func(t *testing.T) {
 					cc.MaybeSkip(t, tc.required)
@@ -120,11 +120,11 @@ func TestWithRealProviders(t *testing.T) {
 		})
 		t.Run("embeddings", func(t *testing.T) {
 			for _, tc := range []realProvidersTestCase{
-				{name: "openai", modelName: "text-embedding-3-small", required: internaltesting.RequiredCredentialOpenAI},
-				{name: "aws-bedrock", modelName: "amazon.titan-embed-text-v2:0", required: internaltesting.RequiredCredentialAWS},
-				{name: "gemini", modelName: "gemini-embedding-001", required: internaltesting.RequiredCredentialGemini},
-				{name: "sambanova", modelName: "E5-Mistral-7B-Instruct", required: internaltesting.RequiredCredentialSambaNova},
-				{name: "deepinfra", modelName: "BAAI/bge-base-en-v1.5", required: internaltesting.RequiredCredentialDeepInfra},
+				{name: "openai", modelName: internaltesting.OpenAIEmbeddingsModelName, required: internaltesting.RequiredCredentialOpenAI},
+				{name: "aws-bedrock", modelName: internaltesting.AWSBedrockEmbeddingsModelName, required: internaltesting.RequiredCredentialAWS},
+				{name: "gemini", modelName: internaltesting.GeminiEmbeddingsModelName, required: internaltesting.RequiredCredentialGemini},
+				{name: "sambanova", modelName: internaltesting.SambaNovaEmbeddingsModelName, required: internaltesting.RequiredCredentialSambaNova},
+				{name: "deepinfra", modelName: internaltesting.DeepInfraEmbeddingsModelName, required: internaltesting.RequiredCredentialDeepInfra},
 			} {
 				t.Run(tc.name, func(t *testing.T) {
 					cc.MaybeSkip(t, tc.required)
@@ -134,8 +134,8 @@ func TestWithRealProviders(t *testing.T) {
 		})
 		t.Run("messages", func(t *testing.T) {
 			for _, tc := range []realProvidersTestCase{
-				{name: "anthropic", modelName: "claude-sonnet-4-5", required: internaltesting.RequiredCredentialAnthropic},
-				{name: "anthropic-aws-bedrock", modelName: "global.anthropic.claude-sonnet-4-5-20250929-v1:0", required: internaltesting.RequiredCredentialAWS},
+				{name: "anthropic", modelName: internaltesting.AnthropicModelName, required: internaltesting.RequiredCredentialAnthropic},
+				{name: "anthropic-aws-bedrock", modelName: internaltesting.AWSBedrockAnthropicGlobalModelName, required: internaltesting.RequiredCredentialAWS},
 			} {
 				t.Run(tc.name, func(t *testing.T) {
 					cc.MaybeSkip(t, tc.required)
@@ -189,8 +189,8 @@ func TestWithRealProviders(t *testing.T) {
 	t.Run("streaming", func(t *testing.T) {
 		client := openai.NewClient(option.WithBaseURL(listenerAddress + "/v1/"))
 		for _, tc := range []realProvidersTestCase{
-			{name: "openai", modelName: "gpt-4o-mini", required: internaltesting.RequiredCredentialOpenAI},
-			{name: "aws-bedrock", modelName: "us.meta.llama3-2-1b-instruct-v1:0", required: internaltesting.RequiredCredentialAWS},
+			{name: "openai", modelName: internaltesting.OpenAIModelName, required: internaltesting.RequiredCredentialOpenAI},
+			{name: "aws-bedrock", modelName: internaltesting.AWSBedrockModelName, required: internaltesting.RequiredCredentialAWS},
 		} {
 			t.Run(tc.name, func(t *testing.T) {
 				cc.MaybeSkip(t, tc.required)
@@ -240,9 +240,9 @@ func TestWithRealProviders(t *testing.T) {
 	t.Run("uses tool in response", func(t *testing.T) {
 		client := openai.NewClient(option.WithBaseURL(listenerAddress+"/v1/"), option.WithMaxRetries(0))
 		for _, tc := range []realProvidersTestCase{
-			{name: "openai", modelName: "gpt-4o-mini", required: internaltesting.RequiredCredentialOpenAI},
-			{name: "aws-bedrock", modelName: "us.anthropic.claude-sonnet-4-5-20250929-v1:0", required: internaltesting.RequiredCredentialAWS},
-			{name: "gemini", modelName: "gemini-3.1-flash-lite", required: internaltesting.RequiredCredentialGemini},
+			{name: "openai", modelName: internaltesting.OpenAIModelName, required: internaltesting.RequiredCredentialOpenAI},
+			{name: "aws-bedrock", modelName: internaltesting.AWSBedrockAnthropicUSModelName, required: internaltesting.RequiredCredentialAWS},
+			{name: "gemini", modelName: internaltesting.GeminiModelName, required: internaltesting.RequiredCredentialGemini},
 		} {
 			t.Run(tc.modelName, func(t *testing.T) {
 				cc.MaybeSkip(t, tc.required)

tests/e2e/basic_test.go

@@ -75,7 +75,7 @@ func Test_Examples_Basic(t *testing.T) {
 
 		for _, tc := range []examplesBasicChatCompletionsTestCase{
 			{name: "openai", modelName: "gpt-4o-mini", skip: !cc.OpenAIValid},
-			{name: "aws", modelName: "us.meta.llama3-2-1b-instruct-v1:0", skip: !cc.AWSValid},
+			{name: "aws", modelName: internaltesting.AWSBedrockModelName, skip: !cc.AWSValid},
 		} {
 			tc.run(t, egSelector)
 		}
@@ -166,7 +166,7 @@ func Test_Examples_Basic(t *testing.T) {
 		})
 
 		// Test Anthropic Messages API (/v1/messages) translated to AWS Bedrock Converse API.
-		// This reuses the existing AWSBedrock route for us.meta.llama3-2-1b-instruct-v1:0 from aws.yaml.
+		// This reuses the existing AWSBedrock route for us.amazon.nova-micro-v1:0 from aws.yaml.
 		t.Run("aws_bedrock_anthropic_messages", func(t *testing.T) {
 			cc.MaybeSkip(t, internaltesting.RequiredCredentialAWS)
 			internaltesting.RequireEventuallyNoError(t, func() error {
@@ -182,7 +182,7 @@ func Test_Examples_Basic(t *testing.T) {
 				)
 
 				msg, err := client.Messages.New(ctx, anthropic.MessageNewParams{
-					Model:     "us.meta.llama3-2-1b-instruct-v1:0",
+					Model:     internaltesting.AWSBedrockModelName,
 					MaxTokens: 256,
 					Messages: []anthropic.MessageParam{
 						anthropic.NewUserMessage(anthropic.NewTextBlock("Say this is a test.")),

tests/e2e/provider_fallback_test.go

@@ -18,6 +18,7 @@ import (
 
 	"github.com/stretchr/testify/require"
 
+	internaltesting "github.com/envoyproxy/ai-gateway/internal/testing"
 	"github.com/envoyproxy/ai-gateway/tests/internal/e2elib"
 )
 
@@ -50,7 +51,7 @@ func Test_Examples_ProviderFallback(t *testing.T) {
 		_ = e2elib.KubectlDeleteManifest(context.Background(), fallbackManifest)
 	})
 
-	const body = `{"model": "us.meta.llama3-2-1b-instruct-v1:0","messages": [{"role": "user", "content": "Say this is a test!"}],"temperature": 0.7}`
+	body := `{"model": "` + internaltesting.AWSBedrockModelName + `","messages": [{"role": "user", "content": "Say this is a test!"}],"temperature": 0.7}`
 
 	// At this point, we haven't configured the fallback for the HTTPRoute generated by the AI Gateway.
 	// So, no matter how many times we try, we should always get a 503 error.