Meta Muse Spark Monitoring with OpenTelemetry

SigNoz published a guide for monitoring Meta's Muse Spark reasoning model family through the OpenAI-compatible Meta Model API using standard OpenTelemetry instrumentation, with the model muse-spark-1.3 priced at $1.25 per million input tokens, $0.15 per million cached input tokens, and $4.25 per million output tokens. The guide states that stock instrumentation drops reasoning tokens, cached input tokens, and time to first chunk, and that cached input is roughly 8x cheaper on the standard tier, causing span-derived cost to overstate real spend by three to five times. It requires Python 3.9 or later, a Meta Model API key, and a SigNoz Cloud account or self-hosted SigNoz instance, and notes that only Chat Completions is traced while the Responses API produces no span.

What is Meta Muse Spark Monitoring? Meta Muse Spark is a reasoning model family served through the Meta Model API. Because the API is OpenAI compatible, you instrument it with the standard OpenTelemetry OpenAI instrumentation rather than a provider specific library, which gives you request traces, model and token usage, latency, and errors. With full Muse Spark monitoring in SigNoz, you can attribute spend to a model and a tier, watch how much of every response is spent on reasoning the caller never sees, measure the wait before the first visible token, and catch the truncated responses that a plain error rate misses. Prerequisites - A SigNoz Cloud account https://signoz.io/teams/ with an active ingestion key, or a self-hosted SigNoz instance https://signoz.io/docs/install/self-host/ - Python 3.9 or later - A Meta Model API key from the Meta AI developer site https://developer.meta.com/ai/ - Network access to api.meta.ai and to SigNoz Monitor Meta Muse Spark with OpenTelemetry Muse Spark is reached over an OpenAI compatible endpoint, so opentelemetry-instrument can trace it with no changes to your application code. 1. Step 1: Install the SDK, the OpenTelemetry distro, and the exporter, then let bootstrap add the matching instrumentation. pip install openai httpx opentelemetry-distro opentelemetry-exporter-otlp opentelemetry-bootstrap --action=install 2. Step 2: Point the OpenAI client at the Meta Model API. Only the base URL and the key change. python from openai import OpenAI client = OpenAI api key="<your-meta-model-api-key ", base url="https://api.meta.ai/v1", response = client.chat.completions.create model="muse-spark-1.3", max tokens=4096, messages= {"role": "user", "content": "Summarize this incident report."} , 3. Step 3: Configure the exporter. export OTEL SERVICE NAME=<service name export OTEL EXPORTER OTLP ENDPOINT=https://ingest.<region .signoz.cloud:443 export OTEL EXPORTER OTLP HEADERS=signoz-ingestion-key=<your-ingestion-key export OTEL EXPORTER OTLP PROTOCOL=http/protobuf export OTEL INSTRUMENTATION GENAI CAPTURE MESSAGE CONTENT=SPAN AND EVENT Verify these values: - <service name : A name for your application, for example muse-chat-api . - <region : Your SigNoz Cloud region https://signoz.io/docs/ingestion/signoz-cloud/overview/ endpoint . - <your-ingestion-key : Your SigNoz ingestion key https://signoz.io/docs/ingestion/signoz-cloud/keys/ . - <your-meta-model-api-key : Your Meta Model API key. 4. Step 4: Run your application under the agent. opentelemetry-instrument python your app.py Each call emits a chat <model span with model, finish reason, and token counts. Only Chat Completions is traced; the Responses API produces no span. Capture Reasoning Tokens, Cache Reads, and Cost The stock instrumentation drops reasoning tokens, cached input tokens, and time to first chunk. Cached input is roughly 8x cheaper on the standard tier, so cost derived from spans alone overstates real spend by three to five times. Wrap your calls in a span that reads them off the response. Setting gen ai.provider.name to meta separates Muse Spark from other providers on the same SDK and stops the wrapper and SDK spans double counting requests. python import time from contextlib import contextmanager from opentelemetry import trace tracer = trace.get tracer "muse-spark" USD per token: input, cached input, output PRICES = { "muse-spark-1.3": 1.25e-6, 0.15e-6, 4.25e-6 , "muse-spark-1.3-contributor": 0.10e-6, 0.002e-6, 0.20e-6 , } @contextmanager def muse call model : with tracer.start as current span f"muse.chat {model}" as span: span.set attribute "gen ai.operation.name", "chat" span.set attribute "gen ai.provider.name", "meta" span.set attribute "gen ai.request.model", model span.set attribute "server.address", "api.meta.ai" call = {} started = time.monotonic yield call response = call.get "response" if response is None or response.usage is None: return usage = response.usage cached = getattr usage.prompt tokens details, "cached tokens", 0 or 0 reasoning = getattr usage.completion tokens details, "reasoning tokens", 0 or 0 fresh = usage.prompt tokens - cached input price, cached price, output price = PRICES model cost = fresh input price + cached cached price + usage.completion tokens output price span.set attribute "gen ai.usage.input tokens", usage.prompt tokens span.set attribute "gen ai.usage.output tokens", usage.completion tokens span.set attribute "gen ai.usage.cache read.input tokens", cached span.set attribute "gen ai.usage.reasoning.output tokens", reasoning span.set attribute "gen ai.cost.total cost", round cost, 8 span.set attribute "gen ai.response.finish reasons", choice.finish reason for choice in response.choices , if call.get "first chunk at" is not None: span.set attribute "gen ai.response.time to first chunk", round call "first chunk at" - started, 3 , Use it around each request: with muse call "muse-spark-1.3" as call: call "response" = client.chat.completions.create model="muse-spark-1.3", max tokens=4096, messages= {"role": "user", "content": "Summarize this incident report."} , For a streaming call, record call "first chunk at" = time.monotonic when the first content chunk arrives, and set stream options={"include usage": True} so the usage record is sent. View Meta Muse Spark Traces in SigNoz Filter the Traces explorer on gen ai.provider.name = 'meta' to see only Muse Spark calls. Opening a trace shows the wrapper span with the SDK span nested under it, and the full attribute set on the right. Meta Muse Spark Monitoring Dashboard Import the Muse Spark dashboard https://signoz.io/docs/dashboards/dashboard-templates/muse-spark-dashboard/ to get cost, token, latency, and reliability panels without building them yourself. Troubleshooting Meta Muse Spark Monitoring No traces appear in SigNoz Confirm the application was launched with opentelemetry-instrument , and that OTEL EXPORTER OTLP ENDPOINT carries your region and the ingestion key header is set. Set OTEL TRACES EXPORTER=console to see what is produced locally. Spans named POST with no gen ai attributes The instrumentor was skipped because httpx is not installed. Run pip install httpx and restart. Responses come back empty gen ai.response.finish reasons is length , meaning max tokens was consumed by reasoning before any visible output. Raise the budget. Errors cannot be told apart The instrumentation sets no error.type , and error spans carry no usage attributes. Read the span status message for the provider error string. Setup OpenTelemetry Collector Optional If you already run an OpenTelemetry Collector https://signoz.io/docs/collection-agents/ , point the application at it with OTEL EXPORTER OTLP ENDPOINT=http://localhost:4318 and configure the Collector's OTLP exporter with your SigNoz endpoint and ingestion key. Related integrations Instrument the other model APIs your team calls, using the same OpenTelemetry pipeline: - Monitor the OpenAI API with OpenTelemetry https://signoz.io/docs/openai-monitoring/ - track token usage, model latency, and error rates across every call - Anthropic monitoring with OpenTelemetry https://signoz.io/docs/anthropic-monitoring/ - trace Claude API calls, usage trends, and failures - Mistral AI observability with OpenTelemetry https://signoz.io/docs/mistral-observability/ - monitor model performance, traces, and token spend - DeepSeek monitoring with OpenTelemetry https://signoz.io/docs/deepseek-monitoring/ - track latency, error rates, and token usage Browse all LLM observability integrations https://signoz.io/docs/llm-observability/ to instrument the rest of your stack.