Best AI Coding Agents
Specialized evaluation focusing on AI coding assistants with detailed scoring breakdown
June 2025 Evaluation
Subset Models & Tests
Usage Note: Any model performing greater than 65% is very usable for day-to-day coding. You'd need to consider cost and time on top of these rankings for if it suits your needs.
Scoring Breakdown
LLM as Judge: Qualitative assessment by Sonnet 3.7 Thinking (0-4 scale)
Deterministic Checks: Automated instruction following verification (0-1 scale)
Diff Errors: Code correctness penalty (negative points for errors)
| Rank | Model | Agent | Final Score | Date |
|---|