Hyperbolic

Hyperbolic sits in a crowded lane — open-model inference clouds — but earns attention two ways: its per-token prices for models like Llama and DeepSeek undercut the closed-model APIs by a wide margin, and it puts those numbers on the page instead of behind a sales call. On top of the OpenAI-compatible inference API, you can rent H100-class GPUs by the hour, so a team can prototype against the hosted endpoint and then rent raw compute for training without changing vendors. Founded by researchers and used by a 250k-strong developer community, it also exposes base-model variants that instruct-only providers hide — handy for anyone doing real experimentation. The trade-offs are the usual ones for a younger challenger: no perpetual free compute tier, a narrower catalog than mega-aggregators, and GPU availability that flexes with demand. But if you want honest, low pricing on open models and the option to rent compute in the same place, Hyperbolic is a solid pick worth benchmarking against Together, Novita and OpenRouter.