This is the layer under everything else: the models the tools in other jobs run on. Billing is by usage, so a monthly figure says nothing, and the filter asks about open weights and about renting GPUs instead.
The open-model repository the whole field links to
The open model in the news is almost always hosted here: this is the field’s shared warehouse, not somebody’s product
Open weights · GPU rental by the hour · Host your own model · Has an API · Single sign-on
In-house video, audio and text models in one place
Runs its own video, speech and text models at once — a rare case of one team holding the full set, with some weights released openly
Many models, one key · Open weights · Phone app · Has an API · Has a free tier
One access point to every provider’s models
Serves every provider’s models through one key with no token markup — you pay the provider’s list price plus a percentage on top-up
Many models, one key · Open weights · Has an API · No free tier · Billed by usage, not subscription
Generative models as a fast API
Specialises in image and video latency — it sells the response time rather than the model, fast enough to work in real time
GPU rental by the hour · Many models, one key · Host your own model · Has an API · No free tier
Open models the community keeps refining
Releases weights freely and trains distributed across other people’s machines — the opposite of the closed labs on this list
Open weights · Host your own model · Runs on your own server · Has an API · Has a free tier
