Global AI MapTop Local AI Leaderboard
Aider Polyglot code-editing results for named models up to 32B parameters. Parameter count is a practical filter, not a measurement of VRAM use or inference speed.
Build snapshot retrieved 2026-10-30T22:39:25.727891+00:00. Live data loads in your browser.
| RANK | SYSTEM / MODEL | SCORE | CONTEXT |
|---|---|---|---|
| 01 | Qwen3 32B ↗ | 40.00 | Aider Polyglot, pass rate after two attempts. Model-size filter; not a memory or speed measurement. 2025-05-08 |
| 02 | QwQ-32B + Qwen 2.5 Coder Instruct ↗ | 26.20 | Aider Polyglot, pass rate after two attempts. Model-size filter; not a memory or speed measurement. 2025-03-07 |
| 03 | QwQ-32B ↗ | 20.90 | Aider Polyglot, pass rate after two attempts. Model-size filter; not a memory or speed measurement. 2025-03-06 |
| 04 | Qwen2.5-Coder-32B-Instruct ↗ | 16.40 | Aider Polyglot, pass rate after two attempts. Model-size filter; not a memory or speed measurement. 2024-12-26 |
| 05 | openhands-lm-32b-v0.1 ↗ | 10.20 | Aider Polyglot, pass rate after two attempts. Model-size filter; not a memory or speed measurement. 2025-04-19 |
| 06 | Qwen2.5-Coder-32B-Instruct ↗ | 8.00 | Aider Polyglot, pass rate after two attempts. Model-size filter; not a memory or speed measurement. 2024-12-22 |
| 07 | gemma-3-27b-it ↗ | 4.90 | Aider Polyglot, pass rate after two attempts. Model-size filter; not a memory or speed measurement. 2025-03-15 |
Method, limits & original data
Aider Polyglot code-editing results for named models up to 32B parameters. Parameter count is a practical filter, not a measurement of VRAM use or inference speed.
Rows are ranked within this benchmark. Different hardware, model versions, harnesses and test conditions can change results. A blank or absent score is never treated as zero. Ties retain equal rank in the interactive view.
Additional benchmark references: Arena · Artificial Analysis · Terminal-Bench. Keyed APIs are not exposed in this website.