A practical test for cheaper coding agent setups
The goal is to find a setup that works for daily development without costing hundreds of dollars each month. The comparison checks whether cheaper API models used through Claude Code and OpenRouter can offer similar value when is already being paid for. The test uses a real from work, not a toy problem.
The project includes PHP, a WordPress admin screen, WP-, , an existing architecture, and a Git workflow. The models being tested include , GPT-5.3 Codex, Kimi K2.5, Qwen3 Coder Next, , MiniMax M3, Xiaomi Mimo 2.5 Pro, models, and free OpenRouter models. The comparison measures cost per finished task, time taken, respect for the existing project architecture, how well requirements are followed, and how much human supervision is needed.
It does not measure general knowledge or LeetCode-style coding problems.
Key points
- The aim is to reduce costs while keeping day-to-day development useful.
- Claude Code and OpenRouter are being tested alongside Codex connected to .
- The test uses a real with existing code, database work, commands, and Git workflow.
- The measured items are task cost, speed, architecture fit, requirement follow-through, and supervision needed.
- General knowledge and LeetCode-style coding tests are intentionally excluded.