C

CoreWeave Model Distillation

Platform

by CoreWeave

Managed post-training workflow that trains a smaller open-weight model on examples from a production task with Serverless supervised fine-tuning, scores it against the model already in use on held-out traffic, and moves traffic over gradually; in preview inside CoreWeave Forge.

Find alternatives to CoreWeave Model Distillation

What it does

• Model training

Sources

Pricing

Official documentation · 18 Oct 2026

"Model Distillation isn't billed separately. You pay standard rates for the services the workflow uses, including Serverless SFT at $2.70 per GPU-hour"; no charging model is stated for the preview itself.

Description

Official documentation · 18 Oct 2026

"Train a smaller model for the task you already run in production. Turn real examples into training data, compare candidates against your current model, and move traffic when the results meet your quality bar."; "Train a smaller open-weight model on examples from your task using Serverless supervised fine-tuning (SFT)."

What it does

Official documentation · 18 Oct 2026

"CoreWeave Model Distillation is a managed post-training capability available through CoreWeave Forge. It uses supervised fine-tuning to train a smaller open-weight model".

Sold within

Official documentation · 18 Oct 2026

"Model Distillation is part of CoreWeave Forge"; "Model Distillation isn't billed separately."

Name

Official documentation · 18 Oct 2026

"CoreWeave Model Distillation".

URL

Official documentation · 18 Oct 2026

The product's own page, titled "CoreWeave Model Distillation" and linked from the Forge menu as "Model Distillation Preview".

Deployment

Official documentation · 18 Oct 2026

"Deploy through CoreWeave Inference and set the share of requests each model receives." CoreWeave manages the training infrastructure.

Intended for

Official documentation · 18 Oct 2026

For teams that already run a frontier model on a production task and want to test "whether a smaller, specialized model can meet your quality requirements".

Also from CoreWeave

• CoreWeave AI Cloud Infrastructure service

  GPU cloud renting NVIDIA accelerators by the hour, with managed Kubernetes, storage and networking for training and inference.

• Mission Control Platform

  Operating layer for CoreWeave Cloud unifying security (IAM, compliance logging), fleet/node lifecycle automation with expert support, and observability (metrics, logs, telemetry), including an AI agent that answers questions about GPU performance and cluster health.

• CoreWeave Sandboxes Infrastructure service

  Managed sandbox compute environments for AI development, billed as an add-on that varies with the buyer's base Weights & Biases plan tier ($10/mo on Free, $25/mo on Pro).

• CoreWeave ARIA AI agent

  AI research agent built into the Weights & Biases platform that analyzes experiment data and production traces, forms hypotheses, designs and runs experiments, and generates dashboards and reports.

• CoreWeave Forge Platform

  CoreWeave Forge provides a development environment for running, observing, improving, and evaluating models and agents.

• CoreWeave Kubernetes Service Infrastructure service

  CoreWeave Kubernetes Service provides managed Kubernetes clusters for building, training, and deploying AI applications.

• CoreWeave SUNK Infrastructure service

  CoreWeave SUNK runs Slurm jobs and Kubernetes workloads on shared compute for AI training.

• SUNK Anywhere Infrastructure service

  SUNK Anywhere licenses SUNK Standard for running Slurm on Kubernetes clusters operated outside CoreWeave.

• CoreWeave Agent Lens Application

  Observability for production AI agents that groups failures across millions of traces, lets teams tune LLM judges and turn each failure into a test case, and checks candidate fixes against real cases before they ship; in preview inside CoreWeave Forge.

• CoreWeave Dedicated Inference API service

  Managed inference for a customer's own or open-source model weights on GPU classes the customer picks, served through OpenAI-compatible vLLM and SGLang endpoints while CoreWeave runs the cluster, routing and scaling, billed per GPU-hour.

• CoreWeave Notebooks Developer tool

  Reactive Python and SQL notebooks that open inside CoreWeave Forge already connected to a project's experiment runs, for analysing training runs and model and agent evaluation results next to the data.

Open CoreWeave in the directory

Something wrong here? Send a correction — quote this product id: coreweave-model-distillation.