Cost governance and savings

Gateway reports spend from every vendor in one place and offers settings that lower it without application changes. To cap spend, use Budgets; for what you’re billed, see Plans and pricing.

Where spend is reported

Usage & spend in the dashboard totals managed and BYOK spend, tokens, and requests across every vendor.

Tab	Breaks spend down by
Overview	Totals over time, managed vs BYOK
By source	Model, provider, or vendor
By attribution	Project, policy, customer, or API key. Policies are labeled by their model alias
Performance	Latency for the same traffic

For a single request, open it in Logs: Estimated cost shows its cost with Merge’s fee underneath, and the summary strip totals both over every request matching your filters. The figures read differently by key type:

Key	Estimated cost	Merge fee
Merge-managed	What you’re billed, with your plan rate applied	The fee’s share of that
Your own (BYOK)	What the tokens cost on your vendor account at list price	What Merge bills you, the only part on your Merge invoice

In your own tooling, read usage.cost from each response; it’s null, not 0, when the serving route has no price on file. Tools that read it automatically, such as Langfuse, need no setup.

Scope traffic to projects by team, product, or environment so By attribution shows who spent what.

Savings levers

Levers stack: routing picks a cheaper model or vendor, while compression and caching shrink what each request sends.

Lever	What it does	Configure it
Intelligent routing on the Cost axis	Scores each prompt’s complexity and pushes most traffic toward the cheaper models in the policy	Routing policies
Priority routing	Puts a cheaper model first and falls back to others only on failure	Routing policies
Tag-based routing	Switches tagged low-stakes traffic, such as internal tools, to a cheaper strategy within the same policy	The project’s Routing tab
Context compression in Cost optimization mode	Compacts JSON and trims long conversations before they reach the model	Configure → Request settings → Defaults, or per project
Prompt caching	Bills a repeated prompt prefix at the vendor’s lower cached rate	Cache markers or session hints on the request

To measure a lever, compare the affected project or policy on By attribution before and after.

Next steps

Budgets

Cap spend at the organization, project, key, or customer scope

Routing policies

Route by cost, performance, or capability across vendors

Context compression

Reduce input tokens on long or JSON-heavy requests