An agent that can read an existing codebase, make changes and open a pull request.
Last updated: 14 Nov 2026·4 tools compared·Not yet tested
🧪
Early picks from research, hands-on tests in progress. We haven't run our hands-on tests on these tools yet, so every tool is marked Not yet tested. Picks are based on current pricing, features and public documentation. How we test
Not a toy demo: a real repository with tests, history and opinions. These agents can read the codebase, plan a change, run commands and hand you a diff or a pull request. All of them now meter usage, so the price you pay depends on how hard you push them.
Our pick: Claude Code. Claude Code is our early pick for working inside an existing repo from the terminal, based on its feature set and how widely it is used for this job. Codex is the closest alternative if you already pay for ChatGPT. We will confirm or overturn this with the hands-on test.
Developers who live in the terminal and want an agent that plans, edits and runs commands across a repo.
Skip it if
You want a free option; Claude Code needs a paid Claude plan or API usage.
Free tier
Not included on the free Claude plan.
Price
Included with Claude Pro ($20/month, or $17/month billed annually) and Max (from $100/month, 5x Pro usage). Usage limits reset every 5 hours plus a weekly cap.
Give each agent the same open-source repo (around 20k lines, with a test suite) and the same three tickets: a bug fix, a small feature, and a refactor. Score passing tests, diff quality, how many nudges it needed, wall-clock time, and plan usage consumed.
Every tool gets the same task, scored on result quality, speed, ease of use and price/free tier. Results and the test date will appear here, and the ranking can change once we have them. Full methodology