How Many AI Agents Can a $6/Month 1GB VM Run? They Measured 200 at Once
swarmagent · reddit · 2026-10-23
The author ran controlled experiments on two identical Google e2-micro VMs (1GB RAM, 2 shared CPUs, $6/month), launching agent waves simultaneously on both. Each agent was a real session making real model calls: search a repo, read a file, write a bug report.
Key findings:
- The model wasn't the bottleneck — their own save path was (one lock for every save)
- After fixes: 100 agents finish in 23s (was 42s), 150 in 51s with no swap, 200/200 complete under full load
- Cost records matched the agents' spend exactly after every wave
The author works on open-source Swarm; full write-up and replay linked in comments.
Related event: 200 Concurrent LLM Agents Run on a $6/Month 1GB RAM VM(2 posts)→
More from coding & agent
- Banks keep piloting agents, but the control plane is what gets missed — krishnan · 2026-10-23
- Agent swarms collaborate 10K hours in simulated GitHubs: Gemini-3.8-flash tops, gpt-6-astra worst — RulinShao · 2026-10-23
- LiteLLM v1.107.0 supply-chain compromise: don't pip fix, delete cook.pth and rotate keys — alvelda · 2026-10-23
- Meta trains an 8B model to pick tools better than a 402B Llama-4-Maverick — dair_ai · 2026-10-23
- Coding agents declaring false 'Done' — a Reddit experiment adds an independent controller to verify completion — aldi949 · 2026-10-23
- LangChain Releases All Interrupt 2026 Agent Conference Recordings — LangChain · 2026-10-23