Agent leaderboards / All sectors / Agent sandboxes

Agent sandboxes: which products coding agents choose

E2B wins about 44% of runs and Modal about 25%.

520 runs5 apps5 agents2 personasupdated 2026-10-07

The interactive board, open on agent sandboxes. Open it full page · this page as Markdown

Read this leaderboard as textrankings, key learnings, method

Key learnings

We asked five coding agents to add a code sandbox to five codebases. That made 520 runs, each asked in several wordings and as two different people. Behind the top pair, Vercel Sandbox took about 16% and Daytona about 14%.

91 of 170

One agent put a different product on top

Codex chose Modal in 91 of its 170 runs. The other four agents all led with E2B, including Cursor at 100 wins.

31 of 56

The persona changed the winner

Asked as an enterprise team, the agents picked Daytona 31 times out of 56 runs. Asked as a senior engineer, they picked E2B far more often than anything else.

20 of 25

Rewording the ask changed the answer

A case is one codebase with one agent, asked the same thing in different words. Of 25 cases, 20 did not come out the same every time.

0 of 165

Some products got named but never chosen

AWS Lambda came up 165 times and won nothing. Cloudflare Workers was named 127 times and won twice. Runloop Devboxes was named 125 times and won three times.

  • The simulated user approved all 520 plans, and sent the agent back in only two runs.
  • Firecracker was named 378 times, but it is the isolation layer under several of these products, not a product.
  • All three wins for Anthropic Code Execution came in the Python AI analyst codebase.
Explore every run in the interactive board

The ranking520 runs

ProductWinsShare
1 E2Be2b.dev 227 44%
2 Modalmodal.com 131 25%
3 Vercel Sandboxvercel.com 81 16%
4 Daytonadaytona.io 73 14%
5 Anthropic Code Executionanthropic.com 3 1%
6 Runloop Devboxesrunloop.ai 3 1%
7 Cloudflare Workersworkers.cloudflare.com 2 0%

By agent, by persona, by wording

By agent

Codex170 runsModal · 91then Daytona · 49
Cursor136 runsE2B · 100then Vercel Sandbox · 18
Claude Code129 runsE2B · 51then Vercel Sandbox · 40
Muse Code · Muse Spark 1.351 runsE2B · 41then Daytona · 10
Grok Build CLI · Grok 4.734 runsE2B · 23then Vercel Sandbox · 8

By persona

Senior engineer464 runsE2B · 224then Modal · 124
Enterprise team56 runsDaytona · 31then Vercel Sandbox · 15

By what the ask stressed

The plain ask340 runsE2B · 154then Modal · 106
Self-hosting, privacy or residency180 runsE2B · 73then Daytona · 47

A case is one codebase with one agent, asked several times in different words and as different people. 20 of 25 cases did not hold to a single product.

How this was measured

Every number on this page comes from a controlled experiment. We took 5 small applications, asked 5 coding agents (Codex, Cursor, Claude Code, Muse Code (Muse Spark 1.3), Grok Build CLI (Grok 4.7)) to add a code sandbox to each of them, in several wordings and as a senior engineer and enterprise team, and let the agent choose the product. Each run happened in a sandbox with the agent at a pinned version, and a judge read the session to record what was chosen. That is 520 runs. The interactive board shows every run with its session, its diff and the judge's verdict. A simulated user stood in for the owner of the codebase: it read the agent's plan and had to approve it before any code was written; it sent the agent back at least once in 2 runs. Read the methodology and the publications.

If you sell in this sector: what these numbers mean for a vendor.

Open the interactive boardThis page as Markdown

Other sectors