docs(responses-api-proxy): refresh Azure example for v1 Responses API

codex-rs/app-server/src/models.rs

@@ -228,7 +228,7 @@ mod tests {
         let models =
             fallback_supported_models_for_provider("cerebras", /*include_hidden*/ false);
 
-        assert_eq!(models.len(), 2);
+        assert_eq!(models.len(), 3);
         assert_eq!(models[0].model, "gpt-oss-120b");
         assert_eq!(models[0].display_name, "OpenAI GPT OSS 120B");
         assert!(models[0].is_default);
@@ -250,6 +250,9 @@ mod tests {
         assert_eq!(models[1].display_name, "Z.ai GLM 4.7");
         assert!(!models[1].is_default);
         assert_eq!(models[1].default_reasoning_effort, ReasoningEffort::Medium);
+        assert_eq!(models[2].model, "gemma-4-31b");
+        assert_eq!(models[2].display_name, "Gemma 4 31B");
+        assert!(!models[2].is_default);
     }
 
     #[test]
@@ -277,16 +280,17 @@ mod tests {
         let models =
             fallback_supported_models_for_provider("anthropic", /*include_hidden*/ false);
 
-        assert_eq!(models.len(), 4);
+        assert_eq!(models.len(), 5);
         assert_eq!(models[0].model, "claude-fable-5");
         assert_eq!(models[0].display_name, "Claude Fable 5");
         assert!(models[0].is_default);
         assert!(!models[0].hidden);
         assert_eq!(models[0].default_reasoning_effort, ReasoningEffort::None);
         assert_eq!(models[0].supported_reasoning_efforts, Vec::new());
         assert_eq!(models[1].model, "claude-opus-4-8");
-        assert_eq!(models[2].model, "claude-sonnet-4-6");
-        assert_eq!(models[3].model, "claude-haiku-4-5-20251001");
+        assert_eq!(models[2].model, "claude-sonnet-5");
+        assert_eq!(models[3].model, "claude-sonnet-4-6");
+        assert_eq!(models[4].model, "claude-haiku-4-5-20251001");
     }
 
     #[test]

codex-rs/app-server/tests/suite/v2/model_list.rs

@@ -562,7 +562,11 @@ wire_api = "responses"
         data.iter()
             .map(|model| (model.id.as_str(), model.is_default))
             .collect::<Vec<_>>(),
-        vec![("gpt-oss-120b", true), ("zai-glm-4.7", false)]
+        vec![
+            ("gpt-oss-120b", true),
+            ("zai-glm-4.7", false),
+            ("gemma-4-31b", false),
+        ]
     );
     assert!(next_cursor.is_none());
     Ok(())

codex-rs/codex-api/src/endpoint/models.rs

@@ -1294,7 +1294,7 @@ mod tests {
         assert_eq!(deepseek.display_name, "DeepSeek V4 Flash");
         assert_eq!(deepseek.context_window, Some(1_048_576));
         assert_eq!(deepseek.max_context_window, Some(1_048_576));
-        assert_eq!(deepseek.experimental_supported_tools, Vec::<String>::new());
+        assert_eq!(deepseek.experimental_supported_tools, vec!["tools"]);
         assert!(!deepseek.supports_parallel_tool_calls);
         let glm = &models[4];
         assert_eq!(glm.display_name, "Z.ai GLM 5.1");

codex-rs/model-provider-info/src/model_provider_info_tests.rs

@@ -69,6 +69,41 @@ query_params = { api-version = "2025-04-01-preview" }
     assert_eq!(expected_provider, provider);
 }
 
+#[test]
+fn test_deserialize_azure_v1_model_provider_toml() {
+    // Azure's v1 API uses the `/openai/v1` base path and no longer needs a
+    // dated `api-version` query parameter. This must deserialize without any
+    // `query_params` while leaving the legacy dated form (asserted in
+    // `test_deserialize_azure_model_provider_toml`) intact.
+    let azure_provider_toml = r#"
+name = "Azure v1"
+base_url = "https://xxxxx.openai.azure.com/openai/v1"
+env_key = "AZURE_OPENAI_API_KEY"
+        "#;
+    let expected_provider = ModelProviderInfo {
+        name: "Azure v1".into(),
+        base_url: Some("https://xxxxx.openai.azure.com/openai/v1".into()),
+        env_key: Some("AZURE_OPENAI_API_KEY".into()),
+        env_key_instructions: None,
+        experimental_bearer_token: None,
+        auth: None,
+        aws: None,
+        wire_api: WireApi::Responses,
+        query_params: None,
+        http_headers: None,
+        env_http_headers: None,
+        request_max_retries: None,
+        stream_max_retries: None,
+        stream_idle_timeout_ms: None,
+        websocket_connect_timeout_ms: None,
+        requires_openai_auth: false,
+        supports_websockets: false,
+    };
+
+    let provider: ModelProviderInfo = toml::from_str(azure_provider_toml).unwrap();
+    assert_eq!(expected_provider, provider);
+}
+
 #[test]
 fn test_deserialize_example_model_provider_toml() {
     let azure_provider_toml = r#"

codex-rs/models-manager/src/model_info_tests.rs

@@ -109,7 +109,7 @@ fn known_nvidia_deepseek_model_uses_local_metadata() {
     assert_eq!(model.display_name, "DeepSeek V4 Flash");
     assert_eq!(model.context_window, Some(1_048_576));
     assert_eq!(model.max_context_window, Some(1_048_576));
-    assert_eq!(model.experimental_supported_tools, Vec::<String>::new());
+    assert_eq!(model.experimental_supported_tools, vec!["tools"]);
     assert!(!model.supports_parallel_tool_calls);
     assert!(!model.used_fallback_model_metadata);
 }

codex-rs/responses-api-proxy/README.md

@@ -79,15 +79,22 @@ codex-responses-api-proxy [--port <PORT>] [--server-info <FILE>] [--http-shutdow
 - `--dump-dir <DIR>`: If set, writes one request JSON file and one response JSON file per accepted proxy call under this directory. Filenames use a shared sequence/timestamp prefix so each pair is easy to correlate.
 - Authentication is fixed to `Authorization: Bearer <key>` to match Codewith expectations.
 
-For Azure, for example (ensure your deployment accepts `Authorization: Bearer <key>`):
+For Azure OpenAI, point `--upstream-url` at the current v1 Responses endpoint. Microsoft's v1 API uses `https://YOUR-RESOURCE-NAME.openai.azure.com/openai/v1/responses` and no longer requires a dated `api-version` query parameter:
 
 ```shell
 printenv AZURE_OPENAI_API_KEY | env -u AZURE_OPENAI_API_KEY codex-responses-api-proxy \
   --http-shutdown \
   --server-info /tmp/server-info.json \
-  --upstream-url "https://YOUR_PROJECT_NAME.openai.azure.com/openai/deployments/YOUR_DEPLOYMENT/responses?api-version=2025-04-01-preview"
+  --upstream-url "https://YOUR-RESOURCE-NAME.openai.azure.com/openai/v1/responses"
 ```
 
+Because the proxy always injects a fixed `Authorization: Bearer <key>` header (see the CLI note above), the value you pipe on `stdin` must be something Azure accepts as a bearer token:
+
+- **API key:** Azure's v1 API accepts the resource API key as a bearer token — this is how the standard `OpenAI()` client authenticates against Azure with key-based auth — so piping `AZURE_OPENAI_API_KEY` works directly. The alternative Azure `api-key: <key>` header form is not produced by this proxy.
+- **Microsoft Entra ID (recommended):** pipe a valid Entra ID access token (for example from `az account get-access-token --resource https://ai.azure.com` or a `DefaultAzureCredential` token provider) instead of the API key. Entra tokens are short-lived, so restart the proxy once the token expires.
+
+The incoming proxy request must still be `POST /v1/responses` with no query string, but `--upstream-url` accepts a full URL. A legacy dated endpoint such as `https://YOUR_PROJECT_NAME.openai.azure.com/openai/deployments/YOUR_DEPLOYMENT/responses?api-version=2025-04-01-preview` therefore keeps working if your resource has not migrated to the v1 path. The same applies to custom `model_providers` entries that set `query_params = { api-version = "..." }`; those dated configurations remain supported.
+
 ## Notes
 
 - Only `POST /v1/responses` is permitted. No query strings are allowed.