Training data, on demand.
Choose a model and environment. Generate the samples your next model needs.
Configure dataset EARLY ACCESS- Model
- Pinned public revision
- Environment
- Math
- Sampling
- Set per request
- Output
- Exported delivery files
An example configuration, not an active order or a delivered dataset.
A dataset you can inspect.
Approved orders expose delivery files and recorded generation details. Inspect and export the results. Validated output in a compatible training format can be saved to your private Datasets library, then selected for an approved training run.
From configuration to next step.
- 01
Choose the task
Select a public model revision and a supported generation environment. Execution still requires qualification.
- 02
Set the scope
Define prompt count, samples per prompt, token limits and sampling. Review an available quote before confirming anything.
- 03
Inspect the delivery
Follow an approved order and export its results. Save validated, compatible output to Datasets before preparing a training run. Other output remains available to inspect and export.
Built around your workload.
No fixed public price is advertised. An eligible configuration may receive a scoped quote in the workspace; estimates are not delivery guarantees.
- Model identity
- Public model repository and pinned revision, subject to runtime compatibility and qualification.
- Task environments
- Code, math, instruction following and logic. The workspace catalog determines which configurations are eligible.
- Generation controls
- Prompt count, samples per prompt, output-token limits, thinking mode where supported, and sampling parameters.
Explore now. Execution is gated.
Generation requires approved pilot access and a qualified, available configuration. The workspace checks ordering eligibility. Browsing, signing in or requesting access does not place an order.