Blog for developers
How to run coding agents on open models, what each step costs at real catalog prices, and where those models fall short of Claude. Every article ends with a setup you can paste and a number you can check. Short configuration references for each tool are in the integration guides; prices and context windows are in the model catalog.
Aider on Qwen3.8 27B: Setup, Temperature 0 and What a Fix Cost
Run Aider on Qwen3.8 27B through an OpenAI-compatible API: environment, model metadata, why temperature 0 costs money, and a measured task in whole and diff mode.
Kilo Code CLI on Qwen3.8 27B: Setup, Config Trap and a Measured Task
Run the Kilo Code CLI on Qwen3.8 27B over an OpenAI-compatible API: where the provider config must live, the compaction threshold, and a measured task.
OpenCode on Qwen3.8 27B: Custom Provider, Costs and a Measured Task
Add Qwen3.8 27B to OpenCode as a custom provider, make its cost display match your bill with a cache price, set the compaction limit, and run a measured task.
Cline on Qwen3.8 27B: Setup, a Measured Task and What It Cost
Set up Cline with Qwen3.8 27B over an OpenAI-compatible API in the CLI and in VS Code, then a measured task: 8 model calls, $0.04, and where the money went.
What an Agent Step Costs: Qwen3.8 27B vs MiMo V2.6 Flash
The cost of one coding-agent step on Qwen3.8 27B and MiMo V2.6 Flash: the formula, a table by context size and cache rate, and steps measured on an open GitHub issue.
Run Claude Code on Open Models for About a Cent a Step
Point Claude Code at Qwen3.8 27B through an Anthropic-compatible API: the setup, what an agent step costs with the prompt cache, and where Claude still does better.