Ponder Inference
Serve the models you train and modify. Connect them to the applications you build.
Saved model
Trained or modified
Exact version
Deployment
Compatible runtime
Reviewed activation
Your app
Requests and responses
Model API
Choose the version. Review the deployment. Serve it through your API.
Review costs before you serve.
- Model serving
- Serving time
- Rate reviewed before activation
- Pay for deployment time, including disclosed startup and idle time, with no additional per-token fee. Review the offered rate before activation.
- Spending controls
- Your budget
- Review the deployment terms
- Inspect available spending limits and stop serving from Console when you no longer need the deployment.
- Retained models
- Storage
- Billed separately from serving
- Keep saved model versions for future use. Review their storage charges separately from active serving capacity.
From saved model to application.
- 01
Choose your model
Start with a saved version from training or a model modification. Keep its configuration, evaluation results, and lineage connected to the deployment.
- 02
Review and activate
Choose a compatible serving configuration. Review the offered rate, startup and idle-time charges, and spending controls before activating capacity.
- 03
Connect your application
Wait for the deployment to be ready, then use the endpoint and model identity shown in Console. Follow requests and usage alongside your model.
Your model.
Ready for real requests.
Keep the behavior you developed connected to the application that uses it. Deployment starts from the exact version you choose.
Serve a trained model
Take a saved SFT or reinforcement-learning candidate into a compatible deployment. Keep its training recipe and evaluation evidence available when deciding what to serve.
Serve a modified model
Use a saved modification with its supported serving runtime. Preserve the exact configuration that produced the behavior you evaluated.
Choose what goes live.
Keep control as it runs.
Training, saving, and serving are separate decisions. Review a candidate, explicitly activate a deployment, and follow its usage through the same Console.
- Exact model versions
- Keep the saved model identity and modification configuration attached to the deployment.
- Application access
- Find the endpoint, authentication instructions, and model identity under API → Your models.
- Usage and lifecycle
- Inspect serving activity and charges, stop new requests, and review final usage when capacity is released.
Put your model to work.
Start with a saved version. Review its deployment. Connect your application.
Build your next model with Ponder Training.