Solutions
One account, every AI job.
Train on named GPUs, serve models behind one API, keep your files on volumes. Pick the job and see how it runs here.
By job
Twelve jobs people run here. Each with real commands and live prices.
- InstanceModel trainingTrain on named GPUs by the hour, and move to a reserved cluster when the job grows.
- InferenceInferenceCall open and frontier models through one API, or serve your own weights on a GPU.
- CrateFine-tuningAdapt an open model to your data with LoRA, QLoRA or a full fine-tune.
- InstanceRAG pipelinesAnswer from your own documents: embeddings and vectors on a GPU, answers from the API.
- InstanceAI agentsGPUs your agent runs jobs on from Claude Code, Codex or Cursor, and the machine stops when the job ends.
- InferenceImage and video generationGenerate images and video from text through one API, paying for each output.
- InferenceSpeech and audioTurn text into speech and audio into text through one API, priced per use.
- InferenceComputer visionRead images, scans and video frames with vision models, or train your own.
- InferenceLanguage modelsChat, reasoning and code from open-weight LLMs, all through one API.
- InstanceData processingClean, deduplicate and tokenize training data on GPUs, then delete the instance.
- CrateResearch and experimentsOpen Jupyter on a GPU, run the experiment, keep the results, stop the bill.
- InstanceEdge and embedded AITrain on a cloud GPU, then compile and quantize the model for devices.