fix(semconv) Legacy attributes support

packages/opentelemetry-instrumentation-langchain/pyproject.toml

@@ -90,5 +90,3 @@ select = ["E", "F", "W"]
 [tool.uv]
 constraint-dependencies = ["urllib3>=2.6.3", "langgraph-checkpoint>=4.0.0", "pip>=25.3"]
 
-[tool.uv.sources]
-opentelemetry-semantic-conventions-ai = { path = "../opentelemetry-semantic-conventions-ai", editable = true }

packages/opentelemetry-semantic-conventions-ai/opentelemetry/semconv_ai/__init__.py

@@ -140,6 +140,81 @@ class SpanAttributes:
     GEN_AI_WATSONX_MIN_NEW_TOKENS = "llm.watsonx.min_new_tokens"
     GEN_AI_WATSONX_REPETITION_PENALTY = "llm.watsonx.repetition_penalty"
 
+    # -----------------------------------------------------------------------
+    # Legacy LLM_* aliases (old names with original string values)
+    # These constants were removed or renamed in v0.5.0.
+    # Kept so non-migrated instrumentation packages continue to work.
+    # TODO: Remove once all instrumentation packages are migrated.
+    # -----------------------------------------------------------------------
+
+    # Removed from SpanAttributes in v0.5.0 (now in upstream OTel gen_ai_attributes)
+    # TODO: migrate each to its GenAIAttributes.GEN_AI_* counterpart
+    LLM_SYSTEM = "gen_ai.system"
+    LLM_REQUEST_MODEL = "gen_ai.request.model"
+    LLM_REQUEST_MAX_TOKENS = "gen_ai.request.max_tokens"
+    LLM_REQUEST_TEMPERATURE = "gen_ai.request.temperature"
+    LLM_REQUEST_TOP_P = "gen_ai.request.top_p"
+    LLM_PROMPTS = "gen_ai.prompt"
+    LLM_COMPLETIONS = "gen_ai.completion"
+    LLM_RESPONSE_MODEL = "gen_ai.response.model"
+    LLM_USAGE_COMPLETION_TOKENS = "gen_ai.usage.completion_tokens"
+    LLM_USAGE_PROMPT_TOKENS = "gen_ai.usage.prompt_tokens"
+    LLM_TOKEN_TYPE = "gen_ai.token.type"
+    LLM_OPENAI_RESPONSE_SYSTEM_FINGERPRINT = "gen_ai.openai.system_fingerprint"
+    LLM_FREQUENCY_PENALTY = "llm.frequency_penalty"
+    LLM_PRESENCE_PENALTY = "llm.presence_penalty"
+    LLM_TOP_K = "llm.top_k"
+    LLM_CHAT_STOP_SEQUENCES = "llm.chat.stop_sequences"
+    LLM_REQUEST_FUNCTIONS = "llm.request.functions"
+    LLM_REQUEST_TYPE = "llm.request.type"
+
+    # Renamed LLM_* -> GEN_AI_* in v0.5.0 (name AND value changed: llm.* -> gen_ai.*)
+    # TODO: migrate each to its SpanAttributes.GEN_AI_* counterpart
+    LLM_USAGE_TOTAL_TOKENS = "llm.usage.total_tokens"
+    LLM_USAGE_TOKEN_TYPE = "llm.usage.token_type"
+    LLM_USER = "llm.user"
+    LLM_HEADERS = "llm.headers"
+    LLM_IS_STREAMING = "llm.is_streaming"
+    LLM_REQUEST_REPETITION_PENALTY = "llm.request.repetition_penalty"
+    LLM_RESPONSE_FINISH_REASON = "llm.response.finish_reason"
+    LLM_RESPONSE_STOP_REASON = "llm.response.stop_reason"
+    LLM_CONTENT_COMPLETION_CHUNK = "llm.content.completion.chunk"
+    LLM_REQUEST_REASONING_EFFORT = "llm.request.reasoning_effort"
+    LLM_USAGE_REASONING_TOKENS = "llm.usage.reasoning_tokens"
+
+    # Renamed LLM_* -> GEN_AI_* in v0.5.0 (name changed, value UNCHANGED)
+    # TODO: migrate each to its SpanAttributes.GEN_AI_* counterpart
+    LLM_REQUEST_STRUCTURED_OUTPUT_SCHEMA = "gen_ai.request.structured_output_schema"
+    LLM_REQUEST_REASONING_SUMMARY = "gen_ai.request.reasoning_summary"
+    LLM_RESPONSE_REASONING_EFFORT = "gen_ai.response.reasoning_effort"
+
+    # OpenAI (renamed LLM_OPENAI_* -> GEN_AI_OPENAI_* in v0.5.0, value unchanged)
+    # TODO: migrate each to its SpanAttributes.GEN_AI_OPENAI_* counterpart
+    LLM_OPENAI_API_BASE = "gen_ai.openai.api_base"
+    LLM_OPENAI_API_VERSION = "gen_ai.openai.api_version"
+    LLM_OPENAI_API_TYPE = "gen_ai.openai.api_type"
+
+    # Watsonx (renamed LLM_* -> GEN_AI_WATSONX_* in v0.5.0, value unchanged)
+    # TODO: migrate each to its SpanAttributes.GEN_AI_WATSONX_* counterpart
+    LLM_DECODING_METHOD = "llm.watsonx.decoding_method"
+    LLM_RANDOM_SEED = "llm.watsonx.random_seed"
+    LLM_MAX_NEW_TOKENS = "llm.watsonx.max_new_tokens"
+    LLM_MIN_NEW_TOKENS = "llm.watsonx.min_new_tokens"
+    LLM_REPETITION_PENALTY = "llm.watsonx.repetition_penalty"
+
+    # Cache attributes — VALUE changed in v0.5.0 (underscore -> dot separator).
+    # LLM_* aliases: for packages using the old LLM_* constant name.
+    # GEN_AI_*_DEPRECATED: for packages already using GEN_AI_* name but expecting old value.
+    # TODO: migrate to SpanAttributes.GEN_AI_USAGE_CACHE_*_INPUT_TOKENS
+    LLM_USAGE_CACHE_CREATION_INPUT_TOKENS = "gen_ai.usage.cache_creation_input_tokens"
+    LLM_USAGE_CACHE_READ_INPUT_TOKENS = "gen_ai.usage.cache_read_input_tokens"
+    GEN_AI_USAGE_CACHE_CREATION_INPUT_TOKENS_DEPRECATED = (
+        "gen_ai.usage.cache_creation_input_tokens"
+    )
+    GEN_AI_USAGE_CACHE_READ_INPUT_TOKENS_DEPRECATED = (
+        "gen_ai.usage.cache_read_input_tokens"
+    )
+
     # Chroma db
     CHROMADB_ADD_IDS_COUNT = "db.chroma.add.ids_count"
     CHROMADB_ADD_EMBEDDINGS_COUNT = "db.chroma.add.embeddings_count"

packages/opentelemetry-semantic-conventions-ai/opentelemetry/semconv_ai/_testing.py

@@ -85,39 +85,61 @@ def test_gen_ai_openai_api_type(self):
 # ---------------------------------------------------------------------------
 
 
-class TestSpanAttributesOldNamesGone:
-    """Assert that removed LLM_* constants no longer exist on SpanAttributes."""
+class TestSpanAttributesLegacyLLMNamesPresent:
+    """Assert that legacy LLM_* constants are still available for non-migrated packages."""
 
     @pytest.mark.parametrize(
-        "old_name",
+        "legacy_name,expected_value",
         [
-            "LLM_SYSTEM",
-            "LLM_REQUEST_MODEL",
-            "LLM_REQUEST_MAX_TOKENS",
-            "LLM_REQUEST_TEMPERATURE",
-            "LLM_REQUEST_TOP_P",
-            "LLM_PROMPTS",
-            "LLM_COMPLETIONS",
-            "LLM_RESPONSE_MODEL",
-            "LLM_USAGE_COMPLETION_TOKENS",
-            "LLM_USAGE_PROMPT_TOKENS",
-            "LLM_USAGE_CACHE_CREATION_INPUT_TOKENS",
-            "LLM_USAGE_CACHE_READ_INPUT_TOKENS",
-            "LLM_TOKEN_TYPE",
-            "LLM_REQUEST_TYPE",
-            "LLM_FREQUENCY_PENALTY",
-            "LLM_PRESENCE_PENALTY",
-            "LLM_CHAT_STOP_SEQUENCES",
-            "LLM_REQUEST_FUNCTIONS",
-            "LLM_TOP_K",
-            "LLM_OPENAI_RESPONSE_SYSTEM_FINGERPRINT",
+            ("LLM_SYSTEM", "gen_ai.system"),
+            ("LLM_REQUEST_MODEL", "gen_ai.request.model"),
+            ("LLM_REQUEST_MAX_TOKENS", "gen_ai.request.max_tokens"),
+            ("LLM_REQUEST_TEMPERATURE", "gen_ai.request.temperature"),
+            ("LLM_REQUEST_TOP_P", "gen_ai.request.top_p"),
+            ("LLM_PROMPTS", "gen_ai.prompt"),
+            ("LLM_COMPLETIONS", "gen_ai.completion"),
+            ("LLM_RESPONSE_MODEL", "gen_ai.response.model"),
+            ("LLM_USAGE_COMPLETION_TOKENS", "gen_ai.usage.completion_tokens"),
+            ("LLM_USAGE_PROMPT_TOKENS", "gen_ai.usage.prompt_tokens"),
+            ("LLM_USAGE_CACHE_CREATION_INPUT_TOKENS", "gen_ai.usage.cache_creation_input_tokens"),
+            ("LLM_USAGE_CACHE_READ_INPUT_TOKENS", "gen_ai.usage.cache_read_input_tokens"),
+            ("LLM_TOKEN_TYPE", "gen_ai.token.type"),
+            ("LLM_REQUEST_TYPE", "llm.request.type"),
+            ("LLM_FREQUENCY_PENALTY", "llm.frequency_penalty"),
+            ("LLM_PRESENCE_PENALTY", "llm.presence_penalty"),
+            ("LLM_CHAT_STOP_SEQUENCES", "llm.chat.stop_sequences"),
+            ("LLM_REQUEST_FUNCTIONS", "llm.request.functions"),
+            ("LLM_TOP_K", "llm.top_k"),
+            ("LLM_OPENAI_RESPONSE_SYSTEM_FINGERPRINT", "gen_ai.openai.system_fingerprint"),
+            ("LLM_IS_STREAMING", "llm.is_streaming"),
+            ("LLM_USAGE_TOTAL_TOKENS", "llm.usage.total_tokens"),
+            ("LLM_USER", "llm.user"),
+            ("LLM_HEADERS", "llm.headers"),
+            ("LLM_RESPONSE_FINISH_REASON", "llm.response.finish_reason"),
+            ("LLM_RESPONSE_STOP_REASON", "llm.response.stop_reason"),
+            ("LLM_CONTENT_COMPLETION_CHUNK", "llm.content.completion.chunk"),
+            ("LLM_REQUEST_REASONING_EFFORT", "llm.request.reasoning_effort"),
+            ("LLM_USAGE_REASONING_TOKENS", "llm.usage.reasoning_tokens"),
+            ("LLM_USAGE_TOKEN_TYPE", "llm.usage.token_type"),
+            ("LLM_REQUEST_REPETITION_PENALTY", "llm.request.repetition_penalty"),
+            ("LLM_REQUEST_STRUCTURED_OUTPUT_SCHEMA", "gen_ai.request.structured_output_schema"),
+            ("LLM_REQUEST_REASONING_SUMMARY", "gen_ai.request.reasoning_summary"),
+            ("LLM_RESPONSE_REASONING_EFFORT", "gen_ai.response.reasoning_effort"),
+            ("LLM_OPENAI_API_BASE", "gen_ai.openai.api_base"),
+            ("LLM_OPENAI_API_VERSION", "gen_ai.openai.api_version"),
+            ("LLM_OPENAI_API_TYPE", "gen_ai.openai.api_type"),
+            ("LLM_DECODING_METHOD", "llm.watsonx.decoding_method"),
+            ("LLM_RANDOM_SEED", "llm.watsonx.random_seed"),
+            ("LLM_MAX_NEW_TOKENS", "llm.watsonx.max_new_tokens"),
+            ("LLM_MIN_NEW_TOKENS", "llm.watsonx.min_new_tokens"),
+            ("LLM_REPETITION_PENALTY", "llm.watsonx.repetition_penalty"),
         ],
     )
-    def test_old_name_absent(self, old_name):
-        assert not hasattr(SpanAttributes, old_name), (
-            f"SpanAttributes.{old_name} should have been removed. "
-            "Consumers should import from opentelemetry.semconv._incubating.attributes.gen_ai_attributes directly."
+    def test_legacy_name_present_with_old_value(self, legacy_name, expected_value):
+        assert hasattr(SpanAttributes, legacy_name), (
+            f"SpanAttributes.{legacy_name} must exist for backward compatibility."
         )
+        assert getattr(SpanAttributes, legacy_name) == expected_value
 
 
 # ---------------------------------------------------------------------------
@@ -181,14 +203,16 @@ class TestSpanAttributesOldValuesAbsent:
             "gen_ai.usage.cache_creation_input_tokens",  # underscore variant (pre-migration)
         ],
     )
-    def test_old_value_not_in_span_attributes(self, old_value):
+    def test_old_value_not_in_new_span_attributes(self, old_value):
         all_values = {
             name: value
             for name, value in vars(SpanAttributes).items()
             if not name.startswith("_") and isinstance(value, str)
+            and not name.startswith("LLM_")  # exclude legacy aliases
+            and not name.endswith("_DEPRECATED")  # exclude deprecated aliases
         }
         assert old_value not in all_values.values(), (
-            f"Old attribute value {old_value!r} is still present in SpanAttributes. "
+            f"Old attribute value {old_value!r} is still present in a GEN_AI_* SpanAttribute. "
             f"It should have been renamed."
         )
 

packages/opentelemetry-semantic-conventions-ai/opentelemetry/semconv_ai/version.py

@@ -1 +1 @@
-__version__ = "0.5.0"
+__version__ = "0.5.1"

packages/opentelemetry-semantic-conventions-ai/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "opentelemetry-semantic-conventions-ai"
-version = "0.5.0"
+version = "0.5.1"
 description = "OpenTelemetry Semantic Conventions Extension for Large Language Models"
 authors = [
   { name = "Gal Kleinman", email = "gal@traceloop.com" },

packages/sample-app/pyproject.toml

@@ -23,7 +23,7 @@ dependencies = [
   "transformers>=4.46.0,<5",
   "replicate>=0.22.0,<0.23.0",
   "cohere>=5.11.1,<6",
-  "anthropic>=0.37.1,<0.38.0",
+  "anthropic>=0.86.0,<1",
   "google-cloud-aiplatform>=1.81.0,<2",
   "python-dotenv>=1.0.1,<2",
   "langchain>=1.0.0,<2.0.0",

packages/sample-app/sample_app/anthropic_joke_example.py

@@ -3,7 +3,10 @@
 from traceloop.sdk import Traceloop
 from traceloop.sdk.decorators import workflow
 
-Traceloop.init()
+Traceloop.init(
+    app_name="sample-app",
+    disable_batch=True,
+)
 
 
 @workflow(name="pirate_joke_generator")
@@ -17,7 +20,7 @@ def joke_workflow():
                 "content": "Tell me a joke about OpenTelemetry",
             }
         ],
-        model="claude-3-opus-20240229",
+        model="claude-haiku-4-5",
     )
     print(response.content)
     return response

packages/sample-app/sample_app/anthropic_joke_streaming_example.py

@@ -17,7 +17,7 @@ def joke_workflow():
                 "content": "Tell me a joke about OpenTelemetry",
             }
         ],
-        model="claude-3-haiku-20240307",
+        model="claude-haiku-4-5",
         stream=True,
     )
     response_content = ""

packages/sample-app/sample_app/anthropic_structured_outputs_demo.py

@@ -31,7 +31,7 @@ def main():
     }
 
     response = client.beta.messages.create(
-        model="claude-sonnet-4-5-20250929",
+        model="claude-haiku-4-5",
         max_tokens=1024,
         betas=["structured-outputs-2025-11-13"],
         messages=[

packages/sample-app/sample_app/anthropic_vision_base64_example.py

@@ -30,7 +30,7 @@ async def task():
 
     client = anthropic.AsyncAnthropic(api_key=api_key)
     message = await client.messages.create(
-        model="claude-3-5-sonnet-20240620",
+        model="claude-haiku-4-5",
         max_tokens=1024,
         messages=[
             {"role": "user", "content": "You are a sub-system ..."},

packages/sample-app/sample_app/async_anthropic_example.py

@@ -16,7 +16,7 @@ async def generate_joke(self):
         response = await anthropic.messages.create(
             max_tokens=1024,
             messages=[{"role": "user", "content": "Tell me a joke about Donald Trump"}],
-            model="claude-3-haiku-20240307",
+            model="claude-haiku-4-5",
         )
         return response.content[0].text
 
@@ -33,7 +33,7 @@ async def generation_helper(self):
             messages=[
                 {"role": "user", "content": "Tell me a joke about Donald Trump"},
             ],
-            model="claude-3-haiku-20240307",
+            model="claude-haiku-4-5",
         )
 
         return response.content[0].text

packages/sample-app/sample_app/async_anthropic_joke_streaming.py

@@ -16,7 +16,7 @@ async def generate_joke(self):
         response = await anthropic.messages.create(
             max_tokens=1024,
             messages=[{"role": "user", "content": "Tell me a joke about Donald Trump"}],
-            model="claude-3-haiku-20240307",
+            model="claude-haiku-4-5",
             stream=True,
             top_p=0.9,
         )
@@ -39,7 +39,7 @@ async def generation_helper(self):
             messages=[
                 {"role": "user", "content": "Tell me a joke about Donald Trump"},
             ],
-            model="claude-3-haiku-20240307",
+            model="claude-haiku-4-5",
             stream=True,
             top_k=50,
         )

packages/traceloop-sdk/pyproject.toml

@@ -86,7 +86,7 @@ test = [
   "pytest-recording>=0.13.1,<0.14.0",
   "pydantic<3",
   "pytest-asyncio>=0.23.7,<1.4.0",
-  "anthropic>=0.25.2,<0.85.0",
+  "anthropic>=0.86.0,<1",
   "langchain>=1.0.0,<2.0.0",
   "langchain-openai>=1.0.0,<2.0.0",
   "pandas>=1.0.0",
@@ -155,7 +155,6 @@ select = ["E", "F", "W"]
 constraint-dependencies = ["urllib3>=2.6.3", "pip>=25.3"]
 
 [tool.uv.sources]
-opentelemetry-semantic-conventions-ai = { path = "../opentelemetry-semantic-conventions-ai", editable = true }
 opentelemetry-instrumentation-agno = { path = "../opentelemetry-instrumentation-agno", editable = true }
 opentelemetry-instrumentation-alephalpha = { path = "../opentelemetry-instrumentation-alephalpha", editable = true }
 opentelemetry-instrumentation-anthropic = { path = "../opentelemetry-instrumentation-anthropic", editable = true }