opentelemetry-instrumentation-genai-openai gaps vs openinference instrumentation

Ready to Fix

Missing Spans

• None

Missing Attributes & Features

• Capture cached prompt tokens on chat in patch.py and chat_wrappers.py from usage.prompt_tokens_details.cached_tokens (OpenInference: llm.token_count.prompt_details.cache_read, semconv: gen_ai.usage.cache_read.input_tokens) #546
• Capture reasoning tokens on chat in patch.py and chat_wrappers.py from usage.completion_tokens_details.reasoning_tokens (OpenInference: llm.token_count.completion_details.reasoning, semconv: gen_ai.usage.reasoning.output_tokens) #547
• Support multimodal image_url parts in input messages on chat in utils.py (OpenInference: message.contents.*.message_content.image.image_url, semconv: gen_ai.input.messages image parts) #548

Blocked on Semantic Conventions

Missing Spans

• Add responses.cancel span (openai.resources.responses.Responses.cancel)
• Add images.generate span (openai.resources.images.Images.generate)
• Add audio.transcriptions.create span (openai.resources.audio.Transcriptions.create)
• Add audio.translations.create span (openai.resources.audio.Translations.create)
• Add audio.speech.create span (openai.resources.audio.Speech.create)
• Add moderations.create span (openai.resources.moderations.Moderations.create) (https://github.com/open-telemetry/semantic-conventions-genai/pull/427)
• Add batches.create span (openai.resources.batches.Batches.create)
• Add files.create span (openai.resources.files.Files.create)

Missing Attributes & Features

• Capture audio tokens on chat from usage.prompt_tokens_details.audio_tokens and usage.completion_tokens_details.audio_tokens (OpenInference: llm.token_count.*_details.audio) (https://github.com/open-telemetry/semantic-conventions-genai/pull/394)
• Capture input text prompt on embeddings (OpenInference: embedding.embeddings.*.embedding.text) (https://github.com/open-telemetry/semantic-conventions-genai/issues/496)
• Capture output embedding vectors on embeddings (OpenInference: embedding.embeddings.*.embedding.vector) (https://github.com/open-telemetry/semantic-conventions-genai/issues/496)

Gaps for Legacy APIs

• Add beta.assistants.create span (openai.resources.beta.assistants.Assistants.create / AsyncAssistants.create) - maps to GenAI gen_ai.create_agent.client (create_agent) #544
• Add beta.threads.runs.create span (openai.resources.beta.threads.runs.Runs.create / AsyncRuns.create) - maps to GenAI gen_ai.invoke_agent.client (invoke_agent) #545
• Add legacy completions.create span (openai.resources.Completions.create / AsyncCompletions.create) - maps to GenAI gen_ai.inference.client (text_completion)
• Support legacy function_call message parameter and assistant choice on chat in utils.py (OpenInference: message.function_call_name, message.function_call_arguments_json)
• Support legacy functions parameter in request tool definitions on chat in utils.py (OpenInference: llm.tools.*.tool.json_schema)

Treat this as an umbrella issue and first choose one unblocked checklist item. Read the relevant entry points named in the issue—patch.py and chat_wrappers.py for chat usage attributes, or utils.py for multimodal and legacy message handling—then compare the listed OpenInference and semantic-convention mappings. Done means one selected gap is supported and its instrumentation output matches the stated mapping.