100,000
Concurrent Sandboxes.
June 19th.
The most ambitious sandbox scale test ever run. Any provider can participate. All results published openly on June 19th, 2026.
Why this matters
Cold-start benchmarks are useful, but they only tell part of the story. As AI agents, code interpreters, and large-scale testing workloads grow, the real differentiator is scaling performance under extreme load.
We run weekly public benchmarks that measure sandbox performance. The Scale Invitational goes further — it answers the question every production team eventually asks: can you actually handle 100K at once?
No marketing claims. No cherry-picked results. One standardized workload, every participant runs it simultaneously, everything published.
What we're testing
Scale to 100,000 concurrent sandboxes
Every participant must successfully launch and sustain 100K sandboxes simultaneously under production-like conditions.
Cold-start latency — p50, p95, p99
Full latency distribution from request to ready sandbox, measured at scale rather than in isolation.
Success rate
What percentage of the 100K sandboxes came up successfully? Failures count.
Meet the Participants
Six providers. One workload. All results published June 2026.
Serverless cloud platform built for AI and ML workloads, with fast cold starts and GPU support.
Lightweight Linux microVMs designed for running AI-generated code safely and at scale.
Isolated sandbox environments built for fast, secure code execution at any scale.
Container platform with flexible build and deployment primitives for production workloads.
Cloud infrastructure optimized for AI and data-intensive workloads at scale.
Isolated sandbox environments built for fast, secure code execution at any scale.
Event Details
June 19th, 2026
Testing day
100,000 concurrent sandboxes
Same workload, every participant
Open results
Published permanently for the community
- 1
Now
Providers request to join the Scale Invitational on X.
- 2
May 2026
Invitations confirmed. Methodology finalized and published.
- 3
June 19th, 2026
Testing day. All participants run the same workload simultaneously.
- 4
June 2026
Full results, deep-dive analysis, and permanent results pages published.
Built on transparency
No pay-to-play. No cherry-picked results. The same methodology behind our weekly public benchmarks — just at a scale nobody's attempted before.