Coding agents for real repos tier list

Agents that edit an existing codebase.

Preliminary: testing in progress. No tool has been ranked yet. The tiers stay empty until our standardized test is run and the receipts are published. The pool below is unranked and in no particular order.

Tiers

S
Awaiting test results
A
Awaiting test results
B
Awaiting test results
C
Awaiting test results

Unranked pool (candidates, no particular order)

What is the best AI coding agent for an existing codebase?

We haven't picked one yet. Testing is in progress. The candidates we're testing are Claude Code, Cursor, OpenAI Codex, GitHub Copilot, Windsurf. The tiers stay empty until every tool has run the same standardized test and the receipts are published.

How will VibeTierList rank each AI coding agent?

Every tool gets the same task: Add a feature + tests to a fixed sample repo; score: tests pass, turns, cost. The prompt, budget, time limit and scoring sheet are identical for every tool.

Can a AI coding agent pay to rank higher?

No. Payment never buys rank on VibeTierList. Sponsored content, if any, is always labeled and kept separate from the tiers.

The standardized test

Add a feature + tests to a fixed sample repo; score: tests pass, turns, cost.

Same prompt, budget, time limit and scoring sheet for every tool. Read the methodology