38 companies
Loading the map…
Organisations that train general-purpose models from scratch, rather than building products on somebody else's. Nearly everything else in the index leans on this layer, which is why a company here is usually the target of a `powered_by` or `uses` edge from several others. Independent labs sit alongside the research arms of large technology companies and national efforts training in their own languages.
Weights published, or API-only? Licence permissive, or restrictive if they are? After that comes context window, modality support, and where inference is allowed to run. Pricing rarely decides anything at this layer. Availability, licence terms and data-handling commitments usually do.