diff --git a/ecologits/tracers/anthropic_tracer.py b/ecologits/tracers/anthropic_tracer.py
index 067426a5..57f0f839 100644
--- a/ecologits/tracers/anthropic_tracer.py
+++ b/ecologits/tracers/anthropic_tracer.py
@@ -127,12 +127,13 @@ class MessageStreamManager(Generic[MessageStreamT]):
     Re-writing of Anthropic's `MessageStreamManager` with wrapped `MessageStream`
     """
 
-    def __init__(self, api_request: Callable[[], MessageStream]) -> None:
+    def __init__(self, api_request: Callable[[], Any], output_format: Any) -> None:
         self.__api_request = api_request
+        self.__output_format = output_format
 
     def __enter__(self) -> MessageStream:
-        self.__stream = self.__api_request()
-        self.__stream = MessageStream(self.__stream)
+        raw_stream = self.__api_request()
+        self.__stream = MessageStream(raw_stream, output_format=self.__output_format)
         return self.__stream
 
     def __exit__(
@@ -149,12 +150,13 @@ class AsyncMessageStreamManager(Generic[AsyncMessageStreamT]):
     """
     Re-writing of Anthropic's `AsyncMessageStreamManager` with wrapped `AsyncMessageStream`
     """
-    def __init__(self, api_request: Awaitable[AsyncMessageStream]) -> None:
+    def __init__(self, api_request: Awaitable[Any], output_format: Any) -> None:
         self.__api_request = api_request
+        self.__output_format = output_format
 
     async def __aenter__(self) -> AsyncMessageStream:
-        self.__stream = await self.__api_request
-        self.__stream = AsyncMessageStream(self.__stream)
+        raw_stream = await self.__api_request
+        self.__stream = AsyncMessageStream(raw_stream, output_format=self.__output_format)
         return self.__stream
 
     async def __aexit__(
@@ -271,7 +273,9 @@ def anthropic_stream_chat_wrapper(
         A wrapped `MessageStreamManager` with impacts
     """
     response = wrapped(*args, **kwargs)
-    return MessageStreamManager(response._MessageStreamManager__api_request)    # noqa: SLF001
+    api_request = response._MessageStreamManager__api_request  # noqa: SLF001
+    output_format = response._MessageStreamManager__output_format  # noqa: SLF001
+    return MessageStreamManager(api_request, output_format)
 
 
 def anthropic_async_stream_chat_wrapper(
@@ -290,7 +294,9 @@ def anthropic_async_stream_chat_wrapper(
         A wrapped `AsyncMessageStreamManager` with impacts
     """
     response = wrapped(*args, **kwargs)
-    return AsyncMessageStreamManager(response._AsyncMessageStreamManager__api_request)  # noqa: SLF001
+    api_request = response._AsyncMessageStreamManager__api_request  # noqa: SLF001
+    output_format = response._AsyncMessageStreamManager__output_format  # noqa: SLF001
+    return AsyncMessageStreamManager(api_request, output_format)
 
 
 class AnthropicInstrumentor:
diff --git a/ecologits/tracers/huggingface_tracer.py b/ecologits/tracers/huggingface_tracer.py
index 3484a584..37711283 100644
--- a/ecologits/tracers/huggingface_tracer.py
+++ b/ecologits/tracers/huggingface_tracer.py
@@ -13,6 +13,10 @@
 from ecologits.tracers.utils import ImpactsOutput, llm_impacts
 
 PROVIDER = "huggingface_hub"
+HF_INFERENCE_URL_PREFIXES = (
+    "https://api-inference.huggingface.co/models/",
+    "https://router.huggingface.co/hf-inference/models/",
+)
 
 
 @dataclass
@@ -31,6 +35,18 @@ class ChatCompletionStreamOutput(_ChatCompletionStreamOutput):
     impacts: Optional[ImpactsOutput] = None
 
 
+def _resolve_model_name(*values: Optional[str]) -> Optional[str]:
+    for value in values:
+        if value is None:
+            continue
+        for prefix in HF_INFERENCE_URL_PREFIXES:
+            if value.startswith(prefix):
+                return value.removeprefix(prefix).removesuffix("/v1/chat/completions")
+        return value
+
+    return None
+
+
 def huggingface_chat_wrapper(
     wrapped: Callable,
     instance: InferenceClient,
@@ -67,7 +83,10 @@ def huggingface_chat_wrapper_non_stream(
     request_latency = time.perf_counter() - timer_start
     output_tokens = response.usage["completion_tokens"]
     input_tokens = response.usage["prompt_tokens"]
-    model_name = instance.model or kwargs.get("model")
+    model_name = _resolve_model_name(response.model, kwargs.get("model"), instance.model)
+    if model_name is None:
+        return response
+
     impacts = llm_impacts(
         provider=PROVIDER,
         model_name=model_name,
@@ -101,13 +120,18 @@ def huggingface_chat_wrapper_stream(
     encoder = tiktoken.get_encoding("cl100k_base")
     prompt_text = "".join([m["content"] for m in kwargs["messages"]])
     input_tokens = len(encoder.encode(prompt_text))
-    model_name = instance.model or kwargs.get("model")
+    default_model_name = _resolve_model_name(kwargs.get("model"), instance.model)
     timer_start = time.perf_counter()
     stream = wrapped(*args, **kwargs)
     output_tokens = 0
     for chunk in stream:
-        output_tokens += 1 # noqa: SIM113
+        output_tokens += 1
         request_latency = time.perf_counter() - timer_start
+        model_name = _resolve_model_name(getattr(chunk, "model", None), default_model_name)
+        if model_name is None:
+            yield chunk
+            continue
+
         impacts = llm_impacts(
             provider=PROVIDER,
             model_name=model_name,
@@ -169,7 +193,10 @@ async def huggingface_async_chat_wrapper_non_stream(
     request_latency = time.perf_counter() - timer_start
     output_tokens = response.usage["completion_tokens"]
     input_tokens = response.usage["prompt_tokens"]
-    model_name = instance.model or kwargs.get("model")
+    model_name = _resolve_model_name(response.model, kwargs.get("model"), instance.model)
+    if model_name is None:
+        return response
+
     impacts = llm_impacts(
         provider=PROVIDER,
         model_name=model_name,
@@ -203,13 +230,18 @@ async def huggingface_async_chat_wrapper_stream(
     encoder = tiktoken.get_encoding("cl100k_base")
     prompt_text = "".join([m["content"] for m in kwargs["messages"]])
     input_tokens = len(encoder.encode(prompt_text))
-    model_name = instance.model or kwargs.get("model")
+    default_model_name = _resolve_model_name(kwargs.get("model"), instance.model)
     timer_start = time.perf_counter()
     stream = await wrapped(*args, **kwargs)
     output_tokens = 0
     async for chunk in stream:
         output_tokens += 1
         request_latency = time.perf_counter() - timer_start
+        model_name = _resolve_model_name(getattr(chunk, "model", None), default_model_name)
+        if model_name is None:
+            yield chunk
+            continue
+
         impacts = llm_impacts(
             provider=PROVIDER,
             model_name=model_name,
diff --git a/tests/conftest.py b/tests/conftest.py
index 44a98df3..d27791c4 100644
--- a/tests/conftest.py
+++ b/tests/conftest.py
@@ -15,6 +15,7 @@ def environment():
     set_envvar_if_unset("OPENAI_API_KEY", "test-api-key")
     set_envvar_if_unset("CO_API_KEY", "test-api-key")
     set_envvar_if_unset("GOOGLE_API_KEY", "test-api-key")
+    set_envvar_if_unset("HF_TOKEN", "hf_test-token")
     set_envvar_if_unset("AZURE_OPENAI_API_KEY", "test-api-key")
     set_envvar_if_unset("AZURE_OPENAI_ENDPOINT", "https://ecologits-azure-openai.openai.azure.com/openai/deployments/gpt-4o-mini/chat/completions?api-version=2024-08-01-preview")
     set_envvar_if_unset("OPENAI_API_VERSION", "2024-06-01")
diff --git a/tests/test_huggingface_hub.py b/tests/test_huggingface_hub.py
index 2de04c9b..c22c32fe 100644
--- a/tests/test_huggingface_hub.py
+++ b/tests/test_huggingface_hub.py
@@ -1,10 +1,15 @@
 import pytest
 from huggingface_hub import AsyncInferenceClient, InferenceClient
 
+HF_INFERENCE_MODEL_URL = (
+    "https://api-inference.huggingface.co/models/"
+    "meta-llama/Meta-Llama-3-8B-Instruct"
+)
+
 
 @pytest.mark.vcr
 def test_huggingface_hub_chat(tracer_init):
-    client = InferenceClient(model="meta-llama/Meta-Llama-3-8B-Instruct")
+    client = InferenceClient(model=HF_INFERENCE_MODEL_URL)
     response = client.chat_completion(
         messages=[{"role": "user", "content": "Hello World!"}],
         max_tokens=15
@@ -16,7 +21,7 @@ def test_huggingface_hub_chat(tracer_init):
 @pytest.mark.vcr
 @pytest.mark.asyncio
 async def test_huggingface_hub_async_chat(tracer_init):
-    client = AsyncInferenceClient(model="meta-llama/Meta-Llama-3-8B-Instruct")
+    client = AsyncInferenceClient(model=HF_INFERENCE_MODEL_URL)
     response = await client.chat_completion(
         messages=[{"role": "user", "content": "Hello World!"}],
         max_tokens=15
@@ -27,7 +32,7 @@ async def test_huggingface_hub_async_chat(tracer_init):
 
 @pytest.mark.vcr
 def test_huggingface_hub_stream_chat(tracer_init):
-    client = InferenceClient(model="meta-llama/Meta-Llama-3-8B-Instruct")
+    client = InferenceClient(model=HF_INFERENCE_MODEL_URL)
     stream = client.chat_completion(
         messages=[{"role": "user", "content": "Hello World!"}],
         max_tokens=15,
@@ -40,7 +45,7 @@ def test_huggingface_hub_stream_chat(tracer_init):
 @pytest.mark.vcr
 @pytest.mark.asyncio
 async def test_huggingface_hub_async_stream_chat(tracer_init):
-    client = AsyncInferenceClient(model="meta-llama/Meta-Llama-3-8B-Instruct")
+    client = AsyncInferenceClient(model=HF_INFERENCE_MODEL_URL)
     stream = await client.chat_completion(
         messages=[{"role": "user", "content": "Hello World!"}],
         max_tokens=15,