Three Local CLI Coding Agents Tested on RTX 4070: Aider, Qwen Code, OpenCode Compared
A developer benchmarked three CLI coding agents — Aider, Qwen Code, and OpenCode — using the same 12 tasks against a locally hosted Qwen 35B model on an RTX 4070 12GB GPU. Qwen Code required specific configuration to function correctly, needing llama-server with the --jinja flag and a 32,768-token context window due to its system prompt and tool definitions consuming nearly 19,400 tokens before any user input. Aider and OpenCode worked with Ollama out of the box, while Qwen Code failed across multiple configurations until the correct llama-server setup was used. Across the 12 standardized tasks, all three tools handled basic operations successfully, though differences emerged on more complex tasks like multi-file refactoring. The comparison highlights that documentation for local model compatibility can be incomplete, and hardware and context-window constraints are critical factors when running large coding agents locally.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.
Discussion (0)
Log in to join the discussion and vote.
Log in