OPENJET CLOUD
Coding agents in the cloud.
On a rented 3090.
OpenJet's engine and Qwen3.8-27B come inside the image, already configured. Run OpenJet's agent on the machine, or point Claude Code, Codex or OpenCode at it. Up to 4 agents decode together on the one GPU, 424 tok/s in total on short coding prompts. You pay by the hour, with no per-token billing or usage caps, and delete the machine when you're done.
DEPLOY
Pick a provider.
Each link opens a public template on the provider's site. OpenJet doesn't rent, bill for or resell the machine. Their prices and terms apply. Prices checked 10 November 2026.
Verified hosts, most around $0.20. Bandwidth extra. SSH in from the instance page and run openjet-agent, or tunnel port 8080 for Claude Code.
Deploy on Vast.aiCommunity Cloud. Secure Cloud is $0.50. Storage extra. HTTPS endpoint through Runpod's proxy, or SSH in and run openjet-agent.
Deploy on RunpodBefore you deploy.
Is there anything to download?
You don't download or set up anything, and the machine doesn't fetch the model when it starts. The provider does have to pull the image, which is about 18 GB, on any host that doesn't already have it, and then the model loads onto the GPU. On a Vast 3090 that already had the image, the model was ready 62 seconds after the machine started.
How do I connect?
The quickest way: SSH in from the provider's instance page and run openjet-agent. The login message shows the engine status and the commands. To use your own agent, the machine prints an API token in its log, and every request needs it. Use the endpoint and token with Claude Code, Codex, OpenCode or OpenClaw. Connection details for each provider ↗
export ANTHROPIC_BASE_URL=<endpoint>
export ANTHROPIC_AUTH_TOKEN=<token>
export ANTHROPIC_MODEL=Qwen3.8-27B-Q4_K_M
export ANTHROPIC_DEFAULT_HAIKU_MODEL=Qwen3.8-27B-Q4_K_M
claudeHow fast is it?
On our own RTX 3090, one conversation runs at 121 tok/s with thinking on and short prompts, and 96 tok/s with 98K tokens in context. A rented card may differ: hosts vary in power limits and load. Methodology and full results ↗
Is my code private?
Your code doesn't go to a model API, but it does run on the provider's rented machine. Marketplace hosts can be third parties, so check the provider's terms and the machine type you choose. If your code must stay on your own hardware, install OpenJet locally.
What does it cost?
You pay the provider's hourly rate for as long as the machine exists, including idle time, plus any storage and bandwidth they charge. Check the price before you deploy, and stop or delete the machine when you're done. OpenJet may earn a referral credit from the provider when you use these links.