Billing
Pricing & service types
You pay for usage. What each request costs is set by the model the router selects, then adjusted by the service type on the API key. Cheaper service types accept cheaper routes; premium service types pay for speed, accuracy, or multi-step optimization.
How a request is priced
Every routed request is priced in three parts:
- Route cost. The router scores eligible models and providers and picks the cheapest one that satisfies your constraints. That provider cost is the base of the bill.
- Service-type multiplier. The service type on your key adjusts the rate. Standard is the baseline. Saver and Scale trade some speed or immediacy for a lower rate. Turbo, Agent, and Precision pay a premium for latency, multi-step optimization, or a stricter quality floor.
- Exact accounting. Input, output, and cached tokens are metered per request and recorded to a micro-USD ledger, so every charge reconciles back to a specific call and its routing receipt.
You are never charged for the router's own overhead separately. The service-type multiplier is the whole story.
Service-type pricing
You choose a service type when you create an API key, so different workloads can carry different economics without changing your code. Multipliers are relative to Standard.
| Service type | vs Standard | Best for |
|---|---|---|
| Standard | baseline | Smart default. Best quality per dollar for everyday production traffic. |
| Saver | -15% | Cheapest route that still clears a sane quality bar. |
| Scale | -30% | Bulk and batch work with huge inputs, tuned for the lowest unit cost. |
| Turbo | +25% | Snappiest first token for an interactive feel. |
| Agent | +30% | Optimizes the whole multi-step run, not a single call. |
| Precision | +40% | Highest accuracy with a guaranteed quality floor. |
| Custom | you set it | Tune the dimensions yourself and price against your own constraints. |
A worked example
Say the router lands on a model whose optimized route works out to a $2.00 effective rate per 1M tokens on a Standard key. Here is what the same work costs under each service type:
| Service type | Rate / 1M tokens | Difference |
|---|---|---|
| Standard | $2.00 | baseline |
| Saver | $1.70 | -15% |
| Scale | $1.40 | -30% |
| Turbo | $2.50 | +25% |
| Agent | $2.60 | +30% |
| Precision | $2.80 | +40% |
The dollar figure is illustrative. The real base changes with the model the router picks for your request, which is exactly what Saver and Scale drive down and what Precision protects.
What the multiplier is for
The multiplier reflects the routing objective, not a hidden markup on a fixed list price:
- Saver and Scale accept cheaper eligible routes and batch-friendly paths, so the rate goes down.
- Turbo prioritizes low time-to-first-token and fast endpoints, which cost more to reserve.
- Precision enforces a stricter quality floor and can escalate to stronger models, so it carries the highest premium.
- Agent optimizes an entire multi-step run with session and cache affinity, priced against the whole trajectory rather than one call.
- Custom lets advanced teams set their own constraints and weights.
Credits, budgets, and reconciliation
You pre-purchase credits and draw them down as you route. Balances, statements, and usage all run through the micro-USD ledger, so a monthly total reconciles back to individual requests. Set budgets and rate limits per key or per workspace, and choose whether a budget breach alerts, blocks, or degrades to lower-cost routing. Any degradation is visible on the routing receipt, never silent.
Bring your own keys
With BYOK, you route through your own provider accounts where supported, so the provider cost lands on your existing contracts. Neural Router charges for routing and control-plane features rather than reselling the tokens.
Value-based savings
For eligible customers, Neural Router can price around verified savings: measured cost reduction at matched quality on your own traffic, compared against a direct-to-model baseline. This is offered where accounting and baseline measurement are production-ready, and available to design partners today.