Keep your datasets, checkpoints and weights. Pay $0.10 per GB per month, while you keep them.
Storage is one flat meter across everything you keep on Nucleus: the JSONL you upload, every checkpoint a run writes and the weights you export. It counts by the GB-hour, so a file stops costing you when you delete it, and expiry on a checkpoint is one call.
One flat rate
$0.10 per GB per month for datasets, checkpoints and exported weights. One meter and one price.
Metered by the hour
The count is in GB-hours, so you pay for a file only while you keep it. Delete it and the meter stops.
Expiry is the lever
Set a TTL on the checkpoints you will not need again and clear it on the ones you serve from. Expiry is the main lever on storage spend.
Forecast before the bill
The usage API reports storage GB-hours for any date range, grouped by day, project or base model.
What you keep, and what keeping it costs
Everything you keep on Nucleus counts on one GB-hour meter at $0.10 per GB per month.
Datasets are the JSONL files you upload for fine-tuning, evaluation or retrieval, validated line by line before a job can use them. Checkpoints are what a training run writes: a training checkpoint holds the full state you resume from and a sampler checkpoint holds the weights an endpoint serves. Exported weights are a checkpoint converted to a LoRA adapter, merged safetensors or GGUF. All three sit in your account and count on the same GB-hour meter.
Storage at a glance
Datasets
- Uploaded via POST /v1/files as a JSONL file
- The API reports bytes, lines and validation status
- Kept until you delete it
- Permanent deletion via DELETE /v1/files/{id}
- Metered by the GB-hour
Checkpoints
- Written by a run, or by a fine-tuning job as result_model
- The API reports bytes, checkpoint_type, expires_at and visibility
- ttl_seconds set at creation, changed with PATCH, null keeps it
- Permanent deletion or keep_checkpoints=false on close
- Metered by the GB-hour
Exported weights
- Exported via POST .../export as LoRA adapter, merged safetensors or GGUF
- The API reports operation, format and status
- Delivered to a signed URL or a Hugging Face repository
- Metered by the GB-hour
Metering and protection
- $0.10 per GB per month by the GB-hour
- storage_gb_hours in the usage API by day, project or model
- AES-256 at rest and TLS in transit
- Training data never trains anyone else's model
- A key bound to another project cannot read a project's files
Storage
Storage
Datasets, checkpoints and exported weights. Metered by the GB-hour, so you only pay while you keep them.
Pick your path
Weight export
Any checkpoint, any format
Checkpoints
Kept until you delete them
Data residency
Choose your region
Bring your own cloud
The platform in your account
Read on
Keep what you train, at $0.10 per GB per month
Storage is the third meter on the bill. See the full table, or read how expiry and deletion work in the API.