Nodus Compute

Nodus Compute is an AI compute platform for training models, running agent code and calling inference APIs. Users bring their own code, container and data, and the platform handles compute placement, job execution, logs, outputs and recovery from saved checkpoints. It targets researchers and engineers running pre training, post training, LoRA fine tuning, reinforcement learning environments and agent sandboxes with evals and longer tasks. A beta inference API is OpenAI compatible. Workloads can be submitted from coding agents such as Claude Code, Codex and Cursor, so an agent can request a GPU, follow progress and retrieve results. Persistent project files carry over between compute sessions, and work can be picked up through VS Code, Jupyter or SSH connections. Pricing is usage based and published per hour in US dollars. GPU rates include H100 SXM 80 GB at $2.60, A100 80 GB at $1.09, L40S 48 GB at $0.793, H200 SXM at $3.593, RTX 4090 at $0.29, B200 at $5.50 and B300 at $6.943. CPU configurations start at $0.0075 per hour for 1 vCPU with 1 GiB of memory. The pricing page sets these rates beside Modal's published rates for the same hardware, and lists a number of GPU types that Modal does not offer. The company states it is backed by Y Combinator.