Providers & Models
Custom Models & PricingEnterprise
Apply negotiated provider prices to catalog models, and add models served from your own endpoints with their own prices.
Custom models extend the model catalog with entries a server administrator defines. A custom variant copies a catalog offering, such as Claude Sonnet 4.5 on Google Vertex, and gives it the prices from your agreement with the provider, so the costs Multi-Router records match what you pay. A custom endpoint model adds a model the catalog does not carry, served by an endpoint you operate, with its own wire ids and prices.
Custom variants
A custom variant is a copy of one catalog offering, identified by the model, provider and variant, with new prices. The copy keeps the offering's regions, wire ids, capabilities, limits, lifecycle and retirement date, and is added to the model as a new variant on the same provider, so the catalog offering stays available beside it. Because it is a copy, later catalog changes to the source offering, including price changes and retirement, do not change it. A router or model group member that names the custom variant is billed at the negotiated prices, and routers that name the catalog variant continue at list price.
To create a custom variant in the dashboard:
- On the Models page, select Clone with new pricing.
- Under Model, search the catalog by name or slug and select the model.
- Under Offering to copy, select the offering. Each row shows its Vendor, Variant, Regions and List pricing.
- Under Name and prices, enter a name, such as Vertex (Negotiated Pricing), and edit the prices, which are prefilled with the list prices for every region. A price box left empty is not charged.
- Select Create custom variant.
The variant id is derived from the name: letters and digits are kept in lowercase and every other run of characters becomes a hyphen, so Vertex (Negotiated Pricing) becomes vertex-negotiated-pricing. The name is up to 80 characters and contains at least one letter or digit, and the derived id must differ from every catalog and custom variant of the model on that provider. Routers name the custom variant by its id in routing.variant. Any offering that is not retired can be copied, except custom variants and custom endpoint models.
Custom endpoint models
A custom endpoint model is a model the catalog does not carry, such as a fine-tuned or self-hosted model, served by an endpoint that implements the OpenAI Chat Completions API. Routers reach it through a provider key for the Custom endpoint provider, which holds the endpoint's base URL; see Supported Providers. To create one, select New custom model on the Models page and complete the following fields:
| Field | Description |
|---|---|
| Publisher | The publisher name, such as acme, up to 80 characters, stored in lowercase. |
| Name | The display name, up to 80 characters, containing at least one letter or digit. |
| Slug | The model's slug in the form publisher/name, such as acme/sim-echo: up to 120 characters, lowercase, with each part starting with a letter or digit and using letters, digits, periods, hyphens and underscores. The Slug follows the Publisher and Name until it is edited, and it must differ from every catalog and custom model. |
| Context | The context window, entered as a number of tokens or with a suffix, such as 128k or 1m. The maximum output equals the context window. |
| Regions and prices | One or more regions, starting with global. Each region has a name of letters, digits, hyphens and underscores, a Wire id, which is the model name the endpoint receives, and its own prices. |
Custom endpoint models are chat models with text input and output. A router uses one by selecting Use a custom endpoint instead, choosing a Custom endpoint key, and entering the model's slug as the Model, and needs no variant. The router's region selects the wire id and price; when the router has no region, Multi-Router uses the model's only region or its global region, and a model with several regions and no global region requires the router to name one in lowercase. The endpoint handles feature support, so Multi-Router forwards streaming, tools and other request features without checking them against the catalog.
Prices
Prices are set per region in US dollars per million tokens, from 0 to 1,000,000. Input and output prices are required, every other price is optional, and a price left empty is not charged. Multi-Router computes the cost of each request from the following prices:
| Price | Applies to |
|---|---|
| Input | Prompt tokens that are not read from or written to the cache. |
| Output | Completion tokens, including reasoning tokens. |
| Cache read | Prompt tokens read from the prompt cache. |
| Cache write (5 minutes) | Prompt tokens written to the prompt cache with a 5-minute lifetime. |
| Cache write (1 hour) | Prompt tokens written to the prompt cache with a 1-hour lifetime. |
| Long context | Input, output and cache prices that replace the standard prices when the whole prompt, including cache reads and writes, reaches the threshold in tokens. A long context price needs both an input and an output price. |
| Fast service tier | Prices for requests the provider reports it served in fast mode, such as Anthropic's fast mode. |
| Web search | The fee for a fixed number of web search tool calls, shown on the Tools tab when the source offering has one. |
The price editor also holds Batch, Priority and Flex tier prices, prices by input type, per-unit prices and cache storage, which Multi-Router stores and displays; request costs use the prices in the table above. Price changes apply to requests made after the change, and earlier requests keep the cost they were recorded with. The cost of each request appears in Traffic and counts toward budgets. Requests to a custom endpoint for a model that is not registered as a custom endpoint model carry no cost.
Where custom models appear
Custom variants and custom endpoint models appear on the Models page for every user, with a Custom chip in the Kind column. A custom variant is listed as an additional offering of its model, and a custom endpoint model is listed with the vendor Custom. The expanded offering shows its origin, such as Copied from vertex · standard, catalog 5.4.1, by admin@example.com on 14 Sep 2026 · prices edited 20 Sep 2026, or Custom endpoint model, by its creator and date. The router model picker lists custom variants with the other variants of the model, and GET /v1/models lists custom endpoint models for organizations that hold an enabled Custom endpoint key.
Editing and deleting
Edit prices changes the prices of a custom variant or custom endpoint model, and the prices must cover every region of the entry. The name, publisher, slug, provider, variant, context window, regions and wire ids are fixed; to change them, delete the entry and create it again. A custom entry can be deleted only when no router or model group in any organization uses it, and a refused deletion lists up to 20 of the routers and model groups that use it, each with its organization.
Who can manage custom models
Custom models are managed in the dashboard by server administrators and have no /v1 API. Only server administrators see Clone with new pricing, New custom model, Edit prices and Delete. On the hosted service, Multi-Router performs server administration. On a self-hosted installation, any server administrator manages custom models, including the administrator created during installation and accounts an administrator grants the role from the Admin page.
Errors
Routers and requests that use custom endpoint models return the following errors:
| Error | Status | Cause |
|---|---|---|
router_region_required | 400 | A router uses a custom endpoint model with several regions and no global region, and names no region. |
router_region_not_offered | 400 | A router names a region the custom endpoint model does not offer. |
region_not_offered | 400 | A router created or changed with /v1/routers names a region the custom endpoint model does not offer. |
region_not_applicable | 400 | A router on a custom endpoint names a region for a model that is not a custom endpoint model. |
In the dashboard, creating a custom variant whose derived id already exists, or a custom endpoint model whose slug already exists, shows the conflict and leaves the form open to change the name or slug.