How do I set up OpenTelemetry for OpenAI cost monitoring?

Install opentelemetry-instrumentation-openai-v2, wire a TracerProvider with OTLP exporter, call OpenAIInstrumentor().instrument(). Every client call gets token spans.

What are gen_ai semantic conventions?

Standard attrs like gen_ai.usage.input_tokens, output_tokens, model, system—auto-captured for GenAI APIs, flowing to any OTLP backend.

Why track LLM tokens instead of just latency?

Latency ignores cost variance; same 3s request can be $0.002 or $0.40. Tokens = dollars.

☁️ Cloud & Infrastructure

OpenTelemetry Cracks Open the Black Box of LLM Costs

Prod LLM teams keep waiting for bills that sting like a bad surprise party. OpenTelemetry's new GenAI instrumentation turns those hidden token feasts into trackable metrics—no new tools required.

theAIcatchup Apr 09, 2026 3 min read

Dashboard graph of LLM input/output tokens and computed costs from OpenTelemetry spans

⚡ Key Takeaways

OpenTelemetry's GenAI conventions auto-capture input/output tokens per span—no manual parsing. 𝕏
Chained calls and model switches demand aggregate tracking; OTLP delivers it natively. 𝕏
Reasoning models hide 'thinking' tokens—conventions expose them, preventing silent overruns. 𝕏

Published by

theAIcatchup

Ship faster. Build smarter.

#GenAI observability #LLM cost monitoring #LLM costs #OpenTelemetry #token monitoring #token tracking

Worth sharing?

Get the best Developer Tools stories of the week in your inbox — no noise, no spam.

Originally reported by dev.to

⚡ Key Takeaways

The 60-Second TL;DR

theAIcatchup

Share this article

Worth sharing?

Related Stories

Kubernetes Devs Get Zero-Code LLM Observability — toil drops, costs plummet

Why Grafana Cloud and OpenLIT Are Your LLM Production Lifeline

Ramp's AI Spend Tracker: Taming the Token Tsunami Before It Drowns Budgets

Chasing Ghosts: Real Distributed Tracing from Android to Your Database

Stay in the loop