CoreWeave launched Forge on October 14 at Fully Connected 2026, consolidating Weights & Biases, OpenPipe, and Marimo into a single platform.
ARIA and Sandboxes reached general availability; Agent Lens, which CoreWeave states detects 20% more critical failures than a general-purpose frontier LLM benchmark, entered preview. CoreWeave also detailed a shipped workload: Forge automated the speculative decoding loop for Kimi K2.6 by connecting training, evaluation, and serving steps that previously required manual handoffs, and CoreWeave states the result topped Artificial Analysis output speed benchmarks.