CoreWeave Launches Forge Development Layer for Models and Agents

CoreWeave announced CoreWeave Forge on September 30, 2026, describing it as a development layer for teams building and improving models and agents. According to the company, Forge runs the full AI loop of run, observe, curate, improve, evaluate and repeat in one connected environment, and it remains open across the models, frameworks and other clouds teams already use. MasterClass and Canva are already building on the platform. The San Francisco-datelined announcement was shared during Fully Connected, CoreWeave’s AI cloud conference, which the company said brings together more than 4,500 customers, partners, developers and AI leaders.

Unifying a Fragmented Toolchain

CoreWeave said model and agent development has been split across tools from separate vendors, none built to work together. For the researchers and engineers carrying a model or agent through its full lifecycle, the company said that fragmentation carries a cost: a production trace does not feed the next training run, a finished experiment does not inform the next evaluation, and every handoff loses a signal or costs more time.

CoreWeave said Forge addresses this by unifying Weights & Biases Models, post-training expertise from OpenPipe and the open-source marimo notebook project with its own services in one connected environment built for continuous improvement. Training runs, experiment tracking, evaluations and agent traces live where the model or agent actually runs, and CoreWeave Registry keeps every checkpoint and agent configuration versioned.

“As more people build with AI, they’re putting models to work with their own data and workflows. That’s where the gaps between model capability and system performance become clear,” said Chen Goldberg, executive vice president of product and engineering at CoreWeave, in the announcement. Engineering teams, Goldberg said, need to understand those gaps, pick out the signals that matter, improve the next version and check whether the change worked under real operating conditions, all as one connected system.

What Ships With CoreWeave Forge

CoreWeave said Forge is available as of the announcement, with new and expanded capabilities across the AI loop.

CoreWeave ARIA, now generally available, is a coding agent that the company said automatically analyzes large-scale experiment data and agent observability data, surfaces what drove a change, and proposes the next experiments worth running, with the evidence to support each proposal. CoreWeave Sandboxes also reached general availability, providing a fresh, isolated environment for every run, whether agent tool use, reinforcement learning or model evaluation, already connected to the rest of the loop so nothing extra has to be stood up to use it.

CoreWeave said Weights & Biases Models helps teams track and compare tens of thousands of experiments and millions of metrics, using advanced visualizations and automated workflows. Its Notebooks component is a developer experience that lives in the environment the work runs in, so a prototype carries directly into training, evaluation and production rather than requiring a rewrite, with custom visualizations built on marimo.

CoreWeave Agent Lens is a new observability and continuous-improvement tool for production agents that analyzes tens of millions of traces and turns them into insights that drive fixes. CoreWeave said Agent Lens detects 20 percent more critical failures and fixes issues at one-tenth the cost, based on a benchmark against a general-purpose frontier LLM.

According to CoreWeave, its Registry versions every model checkpoint and agent configuration in open, portable formats with clear lineage, so the next attempt can start from any point in a team’s history on any stack. CoreWeave Post-Training delivers serverless reinforcement learning, serverless supervised fine-tuning and model distillation, and the company said its serverless reinforcement learning trains 1.4 times faster at 40 percent lower cost than a self-managed setup.

CoreWeave said Inference is now part of Forge, providing playground and inference deployment access to the latest open-weight models. Dedicated Inference is also adding a capability in preview called RL Rollouts, which hot-loads checkpoints into a live deployment so a training loop can keep running without redeploying.

Editions, Trial and the Weights & Biases Transition

Forge is offered in three editions, Free, Pro and Enterprise, which CoreWeave said let teams start without a procurement cycle and grow into the platform. The company directs teams to the CoreWeave Forge website to sign up for Forge Pro and start a free 30-day trial.

The Forge site states that Weights & Biases is now CoreWeave Forge, and that existing Weights & Biases credentials work there to access projects and data. The site displays Canva, Cohere, MasterClass, Perplexity, Runway, Sunday Robotics and Wayve under a “Trusted by the builders of modern AI” heading.

Nick Patience, vice president and practice lead for AI platforms at the Futurum Group, said platform buyers in this market have long had to weigh how much flexibility they give up in exchange for a connected toolchain. “Closing the loop from production back into training in a single environment, while staying open across models, frameworks, and clouds, is a harder engineering problem than either half alone,” Patience said.

CoreWeave was established in 2017 and completed its public listing on Nasdaq in March 2025.