See what every model call costs
telemetry.dev prices each model call when its span arrives, from the token usage that your instrumentation reports. Follow the cost from the project total to the model, the user, the trace, and the span that spent it.
Your coding agent checks that each model call reports its model and token usage, and adds user and session IDs, so that you can see cost by user and by session.
- LLM cost
- $274+18% from the previous 30 days
- LLM tokens
- 96M+11% from the previous 30 days
- Requests
- 48,210+9% from the previous 30 days
- Active users
- 1,284+6% from the previous 30 days
- support-agentuser_7350 · 41 spans$2.84
- research-reportuser_4821 · 28 spans$0.61
- support-agentuser_1093 · 19 spans$0.47
- summarize-sessionuser_2046 · 6 spans$0.12
Follow the cost to the span that spent it
Each model span carries its tokens and its cost. Traces, sessions, and prompt versions add up what their calls spent.
Find the span that spent it
Each span shows its cost. An agent span splits its cost into input, cache writes, cached input, reasoning, and output.
- support-agent$0.0247
- chat claude-sonnet-5-5$0.0208
- lookup_order—
- chat gpt-6-luna$0.0039
- track_shipment—
Cost of support-agent
- Input
- $0.0074
- Cache write
- $0.0060
- Cached input
- $0.0004
- Reasoning
- $0.0040
- Output
- $0.0069
Stay inside the budget
Set a monthly budget in USD for an environment. The alert projects the month from the spend of the last 24 hours.
See alerts- Spent
- $122
- Projected
- $236
- Budget
- $200
Catch an agent that loops
Get an alert when one agent turn costs more than $2 or runs more than 30 spans.
- support-agent38.2s
- chat claude-sonnet-5-5×14
- lookup_order×13
- check_inventory×12
- 1 more span
Know what each prompt version costs
Compare the cost per call of two prompt versions on real traffic before you move the production label.
See prompt managementKnow when a model has no price
A call without a price shows no cost instead of $0. The Missing prices alert tells you when fewer than 90% of model calls have one.
- Calls with a price
- 84%
- Threshold
- 90%
- Window
- 1 hour
No cost could be computed for
- vllm/acme-llama-3.1-8b214 calls
- ollama/acme-support-v358 calls
Report the usage, and telemetry.dev prices it
The integrations report the model and its token usage on each model span. telemetry.dev prices the call at ingest, from a catalog of thousands of models.
- 01Trace your model calls
- 02Add the user and session IDs
- 03Open Overview, Traces, or Sessions
A cost that you send in gen_ai.usage.cost comes before the catalog price. Cache reads and writes use their own rates when the catalog has them, and reasoning tokens count as output.
import OpenAI from "openai";
import { init, propagateAttributes } from "@telemetry-dev/sdk";
import { wrapOpenAI } from "@telemetry-dev/openai";
init({ serviceName: "support-agent" });
const openai = wrapOpenAI(new OpenAI());
await propagateAttributes({ userId: "user_7350", sessionId: "conv_42" }, () =>
openai.responses.create({ model: "gpt-6-luna", input: question }),
);from openai import OpenAI
import telemetry_dev
from telemetry_dev_openai import wrap_openai
telemetry_dev.init(service_name="support-agent")
client = wrap_openai(OpenAI())
with telemetry_dev.propagate_attributes(user_id="user_7350", session_id="conv_42"):
client.responses.create(model="gpt-6-luna", input=question)span.setAttributes({
"gen_ai.request.model": "gpt-6-luna",
"gen_ai.usage.input_tokens": 1204,
"gen_ai.usage.output_tokens": 96,
"gen_ai.usage.cache_read.input_tokens": 1024,
"user.id": "user_7350",
"gen_ai.conversation.id": "conv_42",
});Frequently asked questions
Still have a question? Contact us
How is cost calculated?
telemetry.dev calculates cost at ingest from reported token usage and its model-pricing catalog. Cache-read and cache-write tokens use separate rates when available. Reasoning tokens are not charged separately. An explicit span cost takes precedence. Without matching model pricing or token usage, computed cost is unavailable.
What happens when a model has no price?
telemetry.dev stores no cost for the call, instead of $0, and the trace total leaves it out. The Missing prices alert tells you when fewer than 90% of the model calls that report token usage have a price, and names the models. To price such a model, send its cost in US dollars in gen_ai.usage.cost.
Can I see cost per user or per customer?
Yes, when your spans carry a user ID, such as user.id or the userId of propagateAttributes. The Overview breakdown shows the cost of the users with the most requests, the Sessions list shows the user and the cost of each session, and you can filter Traces by user.
Can I set my own model prices?
Not in the dashboard. The pricing catalog comes from models.dev, LiteLLM, Vercel AI Gateway, and OpenRouter. If you pay a different rate, or use a private model, send the cost of each call in gen_ai.usage.cost, and telemetry.dev uses it instead of the catalog price.
Can I get cost data out of telemetry.dev?
Yes, as JSON. The telemetry.dev MCP server gives an AI agent the requests, tokens, and cost of a project by model, provider, user, or operation, and the telemetry CLI prints the same summary with telemetry usage --json. There is no CSV export.
Trace your next AI request
Choose an integration, add your project API key, and run an instrumented request.
Start free
30,000 ingestion units a month, one per span, log record, metric data point, or Evals write. No credit card required.
npm install @telemetry-dev/sdkTalk to an engineer
Our team shows you tracing, cost tracking, and privacy controls for the frameworks and model providers that you use.