LLM cost estimation

Openlayer estimates the cost of the LLM calls it traces, so you can see what a request, a session, or a day of production traffic costs without pricing each call yourself. The same token prices cost the calls Openlayer makes on your behalf, for AI features and for LLM-as-a-judge tests.

​

How the estimate is computed

When you trace your AI system with one of Openlayer’s SDK integrations, every LLM call records the prompt and completion tokens it used, along with the provider and model that served it. Openlayer multiplies those token counts by the price it has for that provider and model pair. When Openlayer has no price for the pair, the call is estimated at $0. Add or override the price in workspace settings, and the traces that follow pick it up.

​

Where the default prices come from

Default prices come from the Openlayer LLM Cost Tracker, a public service that publishes per-token pricing for models across providers and refreshes daily. The service is open source: it fetches cost data from sources such as OpenRouter and LiteLLM every day and serves it from a REST API you are free to query yourself, with GET /v1/costs returning the full table. Openlayer pulls the catalog from that service once a day, so the prices in your workspace track the ones published on the site. If a pull fails, the prices already in your workspace stay in place until the next one succeeds.

​

View the prices your workspace uses

Go to Workspace settings → LLM costs, listed after AI features. The table holds the default token prices used to estimate LLM cost, plus any overrides for the workspace. All prices are in USD per 1 million tokens, shown with a /Mtok suffix. The columns are:

• Model — the model name, with its provider underneath.
• Input — the price per 1 million input tokens.
• Output — the same, for output tokens.
• Updated — when an override was last changed. Catalog defaults show an em dash.

An info sign beside an input or output price marks a workspace override: its tooltip gives the catalog default the override replaced, or reads Custom model without a default price for a model that exists only in your workspace. Overrides sort to the top of the table, and the remaining rows sort by provider and then model. Find a row with the Search models… box, or narrow the table by Provider, Pricing, Input, and Output. When nothing matches, the table shows No model costs. Viewing the table only requires access to the workspace. Add model, Edit, and Revert to default require permission to update it.

​

Override a price

Override a price when the catalog value does not match what you pay: Overrides apply to the cost estimates of future traces for that provider and model.

​

Add a model Openlayer does not price

Providers with flexible deployments, such as Azure OpenAI and Bedrock, report a deployment name that may not match any model in the catalog. Click Add model and price it yourself in Add model cost:

• Provider — pick one from the searchable list, or add your own.
• Model — the name as it reaches Openlayer, for example my-gpt4-deployment.
• Input and Output — each labeled $ / 1M tokens.
• Additional token categories, a collapsible section for token types you price separately, such as cached or reasoning tokens.

​

Revert to the default

Revert to default, in the row’s Cost options menu, removes an override. If a default exists for that provider and model, it is restored; otherwise the model drops out of the table. Future traces are estimated with the restored default, or with no price when there is none.