CoreWeave Serverless Training (incl. OpenPipe)
CoreWeave, Inc. · docs.coreweave.com · Last checked:
- Managed fine-tuning API
- Reinforcement fine-tuning
- Serverless training
Company status
Active. CoreWeave is publicly listed on Nasdaq (CRWV). CoreWeave announced a definitive agreement to acquire OpenPipe on September 3, 2025; the OpenPipe platform states it has migrated to Weights & Biases and CoreWeave.
Sources 1, 2, 3, 4, 5, 6 · checked 2026-11-08
Capabilities
- Serverless RL: reinforcement learning with your own reward function, for multi-turn agentic tasks
- Serverless SFT: supervised fine-tuning on input/output examples (public preview)
- Training produces LoRA adapters stored as Weights & Biases Artifacts; adapters can be saved locally
- Every trained checkpoint is hosted on CoreWeave Serverless Inference
- Works with the open-source ART framework from OpenPipe
Sources 1, 2, 3, 4, 5, 6 · checked 2026-11-08
Who it tends to fit
- Teams training agents with reinforcement learning
- Teams already using Weights & Biases for experiment tracking
Sources 1, 2, 3, 4, 5, 6 · checked 2026-11-08
Pricing
Contact vendor (training billed per GPU-hour of active training time; rate listed on CoreWeave's post-training pricing page, not confirmed when checked)
Sources 1, 2, 3, 4, 5, 6 · checked 2026-11-08
Published certifications
- SOC 2 (listed in compliance programs; reports via trust portal)
- ISO 27001 (listed in compliance programs)
- ISO 27017
- ISO 27018
Listed as stated by the vendor. Scope can differ by product or subsidiary; request current reports.
Sources 1, 2, 3, 4, 5, 6 · checked 2026-11-08
Limitations to check
- Serverless SFT is in public preview
- LoRA rank and alpha are fixed per base model and cannot be set
- Training sequence limit of 32,768 tokens for most models
- Audit reports are gated behind the trust portal
Sources 1, 2, 3, 4, 5, 6 · checked 2026-11-08
Sources
- Serverless Training: Serverless RL and Serverless SFT; LoRA adapters stored as W&B Artifacts; checkpoints hosted on Serverless Inferencedocs.coreweave.com/trainingChecked 2026-11-08
- Serverless SFT is in public preview; LoRA rank/alpha fixed per base modeldocs.coreweave.com/products/post-training/serverless-training/sftChecked 2026-11-08
- Billed per GPU-hour prorated by active training time; 32,768-token sequence limit; 5 GB free storagedocs.coreweave.com/products/post-training/serverless-training/usage-limitsChecked 2026-11-08
- Sept 3, 2025 definitive agreement to acquire OpenPipe; CoreWeave listed on Nasdaq (CRWV) in March 2025www.coreweave.com/news/coreweave-to-acquire-openpipe-leader-in-reinforcement-learningChecked 2026-11-08
- OpenPipe site states the platform has migrated to Weights & Biases and CoreWeaveopenpipe.ai/Checked 2026-11-08
- Compliance programs: SOC 2, ISO 27001, ISO 27017, ISO 27018docs.coreweave.com/security/trust-compliance/compliance-programsChecked 2026-11-08
Report a correction: [email protected]. Facts are re-checked at least every 90 days.