Skip to content
The Test Cabinet The Test Cabinet

Add or Update a Model

Overview

Section titled “Overview”

The model catalog is owned by the backend. Any model with at least one recorded run already appears in the Models section; curating one gives it a Test Cabinet display name, aliases, a provider logo, and a description. Curation is an in-app edit that takes effect immediately, with nothing to commit, build, or release.

A gg run of a model runs on the providers of its candidate list, one at a time, with OpenRouter as the gateway and the bill. The list keeps the providers that serve the model at its native quantization, at or below its developer’s prices, with a cache-read price and the parameters the run sends, and not banned on the model’s entry. The developer’s own endpoint comes first when it passes, so a model whose developer endpoint is excluded, by the filters or by the account’s privacy settings, runs on the next candidate. A model with no candidate is not testable, and enqueue refuses it with the reason.

The full walkthrough is Adding or Updating a Model.

Prerequisites

Section titled “Prerequisites”
  • A signed-in account in the web console. See Register and Log In.
  • The model’s provider logo URL on svgl.app, which the backend fetches and sanitizes server-side, and its OpenRouter slug when it has one.

Add a model

Section titled “Add a model”
  1. Sign in and open the Models section.
  2. Click Add model for a blank form. When the model already has runs but no curated entry, open its derived entry and click Add this model to seed the form from that run.
  3. Enter the OpenRouter slug and click Fill from OpenRouter to populate the display name, provider, and description from OpenRouter’s catalog. For a model OpenRouter does not list, fill the fields in by hand.
  4. Check the model’s list price: the uncached input, cached input, and output rates per Mtok, and the date the figures were taken. Fill from OpenRouter seeds the three rates. For a model OpenRouter does not list, enter them from the developer’s pricing page. Save refuses a partial or undated set.
  5. Check the result and adjust the wording. A display name is required. Add one model id per alias, each paired with the harness family it works with (Claude Code, Codex, Antigravity, or Others/OpenRouter) so the run form offers a harness only the slugs it can launch. The svgl logo URL is optional.
  6. Click Save.

Update a model

Section titled “Update a model”

Open the model in the Models section, click Edit, change the display name, aliases and their harness families, provider, logo, description, OpenRouter slug, list price, or provider fields, then Save.

Fill from OpenRouter replaces the name, provider, and description rather than filling only the empty fields, and seeds the three list-price rates, so reach for it when you want OpenRouter’s wording back.

Refresh list prices

Section titled “Refresh list prices”

A run’s comparable cost is priced from the model’s list price. The backend refreshes every curated model’s list price from OpenRouter every 24 hours, and fills a missing one when a run binding the model is first enqueued. A model the catalog has no entry for, or one OpenRouter yields no rate for, is refused at enqueue with the reason named.

To refresh on demand, open the Models section and click Refresh, beside Add model. The page reports how many models were updated, confirmed unchanged, and left as they were. A price entered by hand stands until a refresh resolves a rate for that model from OpenRouter.

Provider fields

Section titled “Provider fields”

The form’s provider fields decide the candidate list:

  • Developer provider, set where OpenRouter’s name for the developer’s endpoint does not match the model id’s author segment (qwen/… served by Alibaba). Its standard endpoint’s rates are the list price whenever it lists a complete set, on every refresh. Without it, a model whose developer OpenRouter spells differently is priced at the model’s own price in OpenRouter’s models listing (the pricing field of its /models entry), a third party’s rate.
  • Quantization filter, switched off for a model whose precision no provider discloses, so that every endpoint passes at whatever level it declares.
  • Native quantization, set where the highest level any endpoint declares is wrong for the model.
  • Price ceiling, per Mtok, used when OpenRouter lists no developer endpoint.
  • Banned providers, removed from every gg run of the model.
  • Unknown-quantization providers, kept despite declaring unknown.

At enqueue the backend reads the model’s endpoints listing, builds the list and stamps it onto the launch with the context window. The model’s Stats tab shows the list the next enqueue would build, or the reason it would refuse. See choosing the providers a gg run uses.

Verify

Section titled “Verify”

The new or updated entry appears in the Models section immediately, and its aliases attribute matching runs to it.

Next steps

Section titled “Next steps”