Introducing the best sandboxe for agent swarms:
Lizard Mini
Your agents spend most of their time waiting for the model. Near-zero CPU, a few MB of RAM. You're still paying for a whole computer per agent.
100 Codex agents, 5 minutes each, 20 runs a day, for a month. 1 vCPU / 2 GB per agent:
> Lizard: $40.50
> @northflank: $167
> @vercel: $212
> @e2b: $414
> @daytonaio: $414
> @modal: $594
Each agent gets its own Firecracker microVM for $0.0081/h, so the same budget runs up to 15x more agents, unlimited starts!
