When we launched CoreWeave ARENA earlier this year, we wanted teams to reach production with fewer unanswered questions. They needed evidence beyond benchmarks and from their own workloads to decide what was ready, what needed work, and whether the platform would hold up at the scale they had planned.
Since then, we’ve worked with teams to fine-tune open models for their own applications, meet demanding inference targets, and resolve bottlenecks along the path from model development to serving. Those engagements have given customers a clearer basis for their production decisions, along with a chance to see how our engineers work alongside them.
Now we’re opening that experience to qualified enterprise teams new to CoreWeave through CoreWeave ARENA Pass. This includes the full capabilities of CoreWeave Forge, so your team can explore ways to run, evaluate, and improve your AI.
Founder Story:
Hear CoreWeave’s Brian Venturo talk about what inspired CoreWeave ARENA.
What teams have achieved using CoreWeave ARENA
You.com came to CoreWeave ARENA with pretraining and inference workloads and questions that went well beyond GPU performance. Developer productivity, platform control, support, and pricing all mattered to the team.
Running an actual workload through the program allowed You.com engineers to review training and evaluation runs while tracking model versioning. Our engineers helped their workload meet its tight tail-latency requirements, and showed how compute quotas scaled in production when demand grew beyond its original allocation.
Running the actual workload through CoreWeave ARENA gave us insights about model-repo integration, quota elasticity, and consolidation savings. All of these factors determine whether the platform holds up in production.
Bojan Babic, Staff Research Engineer, You.com
Other teams have come away with answers to specific launch questions. A voice AI team met its evaluation criteria with 1,200 concurrent sessions and p99 server-side processing latency below 1.3 seconds during a two-week evaluation. A consumer AI team saw its model’s throughput nearly match its own baseline without the serving optimizations it normally used, then chose CoreWeave for production because of those insights.
Periodic Labs stress-tested storage throughput before committing to a training path. 16 nodes uploaded 16 TB in seven minutes, while aggregate reads across approximately 43 nodes reached roughly 300 GB/s. Those measurements gave the team evidence to assess whether storage could keep up with its planned training workloads before making a larger infrastructure commitment.
Scripps Research, an independent nonprofit biomedical research institute, validated Slurm workflows on CoreWeave Kubernetes Service across two research teams, then moved straight into production.
Test the partnership, not just the platform
CoreWeave ARENA lets you test our partnership along with the technology. Through CoreWeave Direct-to-Expert, the engineers who build and run the platform work alongside your team on architecture decisions and the bottlenecks that emerge as you test. You can draw on their experience running AI at scale, rather than having to learn every lesson yourself. You get time and energy back to test ideas and follow unexpected results, including the connections and discoveries AI can help bring to light.
Founder Story:
See how Anam’s Caoimhe Murphy is bringing a human touch to AI avatars.
Know how you’ll make the next version better
An application can meet its latency target and still give an answer you wouldn’t want a customer to see. Your team needs to understand what happened, make a change, and check whether the next version is better in a full production environment.
CoreWeave Forge brings AI development into a connected environment, so what you learn in one step can inform the next. In CoreWeave ARENA, you can test that AI loop on your own application, with our team helping you work through the results.
- Run: Deploy models and agents in production and generate real-world signals.
- Observe: Capture traces, metrics, tool usage, and behavioral feedback from metal to agent to understand performance.
- Curate: Transform production signals into high-value datasets and continuously refreshed evaluation suites.
- Improve: Use curated data to optimize agents, shift models, improve the harness, refine models with various techniques from reinforcement learning to supervised fine-tuning to model distillation, and deliver better quality, faster performance, and lower cost.
- Evaluate: Measure new models and agents against repeatable quality standards before, during, and after deployment.
Test the infrastructure, not just the app
Where your production decision also involves CoreWeave Cloud, we test the infrastructure questions alongside the application. That can mean examining latency and throughput as concurrency rises, comparing Serverless and Dedicated Inference economics, or investigating how a workload behaves when capacity changes. CoreWeave Mission Control helps connect the results to what’s happening underneath.
CoreWeave ARENA helps you work through the questions that matter to your launch and the releases that follow, using the parts of CoreWeave Forge and CoreWeave Cloud relevant to your workload.
Founder Story:
See how Rime’s Lily Clifford is taking a linguist-led approach to voice AI.
Work with us on the decision ahead
We’re prioritizing enterprise teams new to CoreWeave that are working toward a production deployment in the coming months. You might be preparing an agent for customers, scaling an inference service, or fine-tuning a model for a specific use case.
CoreWeave ARENA Pass brings you into a focused engagement with our engineers. We agree on the success criteria, build the evaluation around your workload, and work through the findings together. Qualification helps us establish whether we have the right capabilities and environment for the questions you need answered. You don’t need to arrive with a finished plan.
At the end, you’ll have results tied to your requirements and a recommendation for what comes next. That might be a production deployment, another test to resolve an open question, or a change to the approach before you invest further. We want you to be able to explain that decision to the people who will build, run, and depend on the application.
Tell us what you’re building, when you expect to put it into production, and what you need to prove. We’ll work with you to assess fit and shape the evaluation.
Apply for CoreWeave ARENA Pass
Explore the CoreWeave ARENA program
