Every model in Borg now has its own page

Borg's model directory gives each listed model a page with its specifications, sources, and available gateway measurements. The pages help you see what is known, where a number came from, and which details are not published.

Start with the specification

Each model page has an at-a-glance table with the provider and family, release date and context window where published, input and output modalities, reasoning and thinking modes, open-weights status, knowledge cutoff, Borg lane IDs, and Fusion roles. Missing values are marked as not published rather than guessed.

Lane IDs matter if you want to choose a model directly through the gateway rather than use a Fusion profile. Pages also describe what the model is listed as good for and how Borg uses it. Related links connect provider models, shared Fusion roles, and versions.

A version link does not mean two versions have identical capabilities.

The breadcrumb leads back through the provider to the directory. Check each model page's own facts before comparing versions.

Know where a number comes from

Vendor benchmark rows link to their sources. They are vendor evaluations, not an independent Borg ranking. A benchmark may use its own settings and comparison group, so it is not interchangeable with measured gateway traffic. The directory does not publish Artificial Analysis-origin numbers or fill gaps with estimates.

Where available, the measured-on-Borg section gives median and 95th-percentile time to first token and end-to-end output tokens per second, with sample counts. A request success rate appears only when it passes the publication rule. These measurements describe observed gateway traffic, not a promise about every task.

The page notes explain the measures, and our speed explainer gives more detail. Pricing is separate: pages identify subscription inclusion and link to available vendor API list prices. Those prices do not set a fixed cost for your workload.

Read the same material as text

Model pages also have an llms.txt text counterpart. It carries published details, available gateway metrics, and source references without requiring a reader to extract them from the layout. It is a companion to the HTML page, not a separate claim about model ability.

The source link and its qualifications matter when checking a fact.

The pricing explainer and cost savings calculator help with cost questions. A page's sources section gathers its provider links.

Start at all makers and models, then follow a provider to an individual model. The directory is meant to answer which models are listed, how to select them, and what the available evidence says.