From 98784360e85186f798c7ffac797aba4020c964fe Mon Sep 17 00:00:00 2001
From: Yuneng Jiang <yuneng@berri.ai>
Date: Sat, 5 Sep 2026 10:37:14 -0700
Subject: [PATCH] test(e2e): cover Anthropic /chat/completions streaming and
 tool calls

Adds TestAnthropicChatCompletions to the chat completions regression suite,
registering a claude-haiku-4-5 deployment via /model/new and asserting the
streamed call delivers real content deltas and a tool-forced call returns a
well-formed get_weather tool_call on both the non-streamed and streamed paths.
Covers three P0 registry cells that had no e2e test.
---
 .../test_chat_completions_regression_e2e.py   | 98 ++++++++++++++++++-
 1 file changed, 97 insertions(+), 1 deletion(-)

diff --git a/tests/e2e/llm_translation/test_chat_completions_regression_e2e.py b/tests/e2e/llm_translation/test_chat_completions_regression_e2e.py
index 68c0dfab897..87bd32d8dab 100644
--- a/tests/e2e/llm_translation/test_chat_completions_regression_e2e.py
+++ b/tests/e2e/llm_translation/test_chat_completions_regression_e2e.py
@@ -11,7 +11,7 @@ fails that provider's row here.
 
 The per-provider classes below cover the OpenAI-compatible /chat/completions
 translation for providers customers reach by registering their own deployment
-via /model/new (Cohere, Gemini, hosted_vllm), each deleted on teardown.
+via /model/new (Cohere, Gemini, hosted_vllm, Anthropic), each deleted on teardown.
 """
 
 from __future__ import annotations
@@ -46,6 +46,7 @@ pytestmark = pytest.mark.e2e
 COHERE_BACKEND = "cohere/command-r-08-2024"
 GEMINI_BACKEND = "gemini/gemini-2.5-flash"
 OPENAI_BACKEND = "openai/gpt-5.6"
+ANTHROPIC_BACKEND = "anthropic/claude-haiku-4-5-20251001"
 BEDROCK_CONVERSE_BACKEND = "bedrock/us.anthropic.claude-haiku-4-5-20251001-v1:0"
 
 
@@ -746,3 +747,98 @@ class TestBedrockConverseChatCompletions:
 
         response = unwrap(client.proxy.chat(key, ChatBody(model=model, messages=_vision_messages(), max_tokens=32)))
         _assert_describes_cat(response)
+
+
+class TestAnthropicChatCompletions:
+    """Anthropic via the OpenAI-compatible /chat/completions path, the translation
+    customers on the OpenAI SDK rely on when they route to Claude. The streamed call
+    must deliver real content deltas, and a tool-forced call must come back as a
+    well-formed tool_call on both the non-streamed and streamed paths.
+    """
+
+    def _register(self, client: PassthroughClient, resources: ResourceManager, prefix: str) -> str:
+        model = f"{prefix}-{unique_marker()}"
+        model_id = client.proxy.create_model(
+            model, LiteLLMParamsBody(model=ANTHROPIC_BACKEND, api_key="os.environ/ANTHROPIC_API_KEY")
+        )
+        resources.defer(lambda: client.proxy.delete_model(model_id))
+        return model
+
+    @pytest.mark.covers(
+        "llm.chat_completions.anthropic.basic.stream.works",
+        exercised_on=["chat_completions"],
+    )
+    def test_anthropic_chat_streams_real_content(
+        self, client: PassthroughClient, resources: ResourceManager
+    ) -> None:
+        model = self._register(client, resources, "e2e-anthropic-stream")
+        key = resources.key()
+
+        result = client.proxy.chat_stream(
+            key,
+            ChatBody(
+                model=model,
+                messages=[
+                    ChatMessage(role="user", content=f"Count from 1 to 5, one number per line. {unique_marker()}")
+                ],
+                max_tokens=64,
+                stream=True,
+            ),
+        )
+        _assert_streamed_completion(result)
+
+    @pytest.mark.covers(
+        "llm.chat_completions.anthropic.tool_use.nonstream.works",
+        exercised_on=["chat_completions"],
+    )
+    def test_anthropic_chat_returns_tool_call(
+        self, client: PassthroughClient, resources: ResourceManager
+    ) -> None:
+        model = self._register(client, resources, "e2e-anthropic-tool")
+        key = resources.key()
+
+        response = unwrap(
+            client.proxy.chat(
+                key,
+                ChatBody(
+                    model=model,
+                    messages=[
+                        ChatMessage(role="user", content="What is the weather in San Francisco? Use the get_weather tool.")
+                    ],
+                    tools=[_WEATHER_TOOL],
+                    tool_choice="required",
+                    max_tokens=128,
+                ),
+            )
+        )
+        _assert_weather_tool_call(response)
+
+    @pytest.mark.covers(
+        "llm.chat_completions.anthropic.tool_use.stream.works",
+        exercised_on=["chat_completions"],
+    )
+    def test_anthropic_chat_streams_tool_call(
+        self, client: PassthroughClient, resources: ResourceManager
+    ) -> None:
+        model = self._register(client, resources, "e2e-anthropic-tool-stream")
+        key = resources.key()
+
+        result = client.proxy.chat_stream(
+            key,
+            ChatBody(
+                model=model,
+                messages=[
+                    ChatMessage(role="user", content="What is the weather in San Francisco? Use the get_weather tool.")
+                ],
+                tools=[_WEATHER_TOOL],
+                tool_choice="required",
+                max_tokens=128,
+                stream=True,
+            ),
+        )
+        assert result.ok and result.is_streaming, f"tool stream was not established: {result}"
+        assert result.stream_error is None, f"tool stream carried an error event: {result.stream_error}"
+        name, arguments = _streamed_tool_call(result.stream_events)
+        assert name == "get_weather", f"streamed tool call named {name!r}: {result.stream_events[:5]}"
+        args = _WeatherArgs.model_validate_json(arguments)
+        assert args.location.strip(), f"streamed tool call arguments missing location: {arguments!r}"