Stealth benchmarks · for providers

Run our benchmarks privately.

Daily runs and on-demand reruns. Results stay off the public leaderboard.

Get stealth benchmarks →See pricing

Our benchmark schedule:

Next benchmark run

private

AI Gateway — Anthropic daily

Sun, Nov 8 · 9:00 AM

Benchmark Feed

Nov 6, 2026

Sandbox DAX

Clone → install → typecheck · Lower is better

Sandboxes

  1. 1PromigencePre-release31.67s
  2. 2Miosa34.17s
  3. 3Isorun44.61s
  4. 4Archil45.60s
  5. 5CreateOS52.68s
  6. 6Upstash53.58s
  7. 7Buddy54.46s
  8. 8Microsandbox55.09s
  9. 9Namespace55.29s
  10. 10Mosaic58.07s
  11. 11GravixlayerPre-release58.66s
  12. 12Sail59.16s
  13. 13Superserve59.36s
  14. 14Blaxel61.05s
  15. 15Arker65.53s
  16. 16Tenki67.92s
  17. 17Tensorlake73.07s
  18. 18Daytona81.53s
  19. 19Novita86.85s
  20. 20Vercel87.46s
  21. 21Sandbox094.56s
  22. 22Modal96.33s
  23. 23E2B99.72s
  24. 24Beam99.88s
  25. 25Agent37110.02s
  26. 26Cloudflare182.72s

Sandbox Reliability

Daily uptime · Higher is better

Reliability

  1. 1Agent37100%
  2. 2Arker100%
  3. 3Beam100%
  4. 4Blaxel100%
  5. 5Buddy100%
  6. 6Google Cloud Run100%
  7. 7Cloudflare100%
  8. 8CodeSandbox100%
  9. 9CreateOS100%
  10. 10Daytona100%
  11. 11E2B100%
  12. 12Freestyle100%
  13. 13sfcompute100%
  14. 14Isorun100%
  15. 15Microsandbox100%
  16. 16Miosa100%
  17. 17Modal100%
  18. 18Namespace100%
  19. 19Novita100%
  20. 20Run Cloud100%
  21. 21Sail100%
  22. 22Sandbox0100%
  23. 23Sprites100%
  24. 24Superserve100%
  25. 25Tenki100%
  26. 26Tensorlake100%
  27. 27Upstash100%
  28. 28Vercel100%
  29. 29Mosaic99.861%
  30. 30Archil99.791%
  31. 31GravixlayerPre-release93.819%
  32. 32PromigencePre-release93.819%

What ComputeSDK does

Benchmarks

ComputeSDK measures sandbox, storage, browser, and AI gateway providers against the same workloads and publishes the rankings free every week. Every workload and raw result is in the open methodology.

Built for agents

Create sandboxes, run commands, and read benchmark results over the REST API (OpenAPI) or the MCP server — llms.txt lists every machine-readable surface.

For providers

Stealth benchmarks run the public suite daily and privately for your provider, and the Market sells your idle capacity at your own price.