rent a GPU for a coding agent
Rent a GPU for your coding agent, already set up
Renting a 24 GB GPU is cheap. Setting up a fast model server on it is the slow part, and the template does it for you.
OpenJet's public templates on Vast.ai and Runpod start an RTX 3090 with OpenJet's engine and Qwen3.8-27B installed. There's no CUDA setup, no model download to manage and no server to configure: deploy, then connect your agent.
Why local wins
- Preinstalled. The image contains OpenJet's engine and Qwen3.8-27B Q4_K_M, configured for the RTX 3090.
- Twice llama.cpp's speed. 121 tok/s against llama.cpp's 61 on short prompts, and 96 against 37 with 98K tokens in context, on an RTX 3090.
- Your account, your bill. You rent on your own Vast.ai or Runpod account and pay them directly: Vast verified hosts run about $0.12 to $0.37 an hour, Runpod Community Cloud $0.22.
Common deployment paths
- Try a 27B coding model without buying a GPU.
- Spin up a GPU for a day of agent work and delete it after.
- Give a small team a shared model endpoint for a project.
Frequently asked questions
- Which GPU does the template use?
- A single RTX 3090 with 24 GB, which fits Qwen3.8-27B with about 100K tokens of context.
- Do I pay while the machine is idle?
- Yes. Providers bill by the hour while the machine exists, so stop or delete it when you're done.
Explore OpenJet
Embed a local model with the Python SDK, run Qwen3.8-27B as a local coding agent, compare local coding-agent alternatives, read the docs, or go back to the homepage.