GreenCode - EffiBench Correctness default: leaderboard
GreenCode paper v1: EffiLearner five-epoch self-refinement with execution and memory profiler feedback on all 1,000 EffiBench problems (100 tests each). Pass@1 is functional correctness, not energy efficiency; native percentage. Default greedy decoding, native context and 2,048 output tokens; each variant changes only the named setting. Unspecified Ministral/Nemotron release generations retain source-qualified identities. default. Metric: Pass@1 (%).
Metric: Pass@1 (%). Source: arxiv.org. Saturation forecast: Around January 2027. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 27B | 81 | #189 |
| 2 | Qwen 3.5 35B A3B | 78.6 | #261 |
| 3 | Qwen 3.5 9B | 66.8 | #392 |
| 4 | Qwen 3.5 4B | 51.3 | #502 |
| 5 | GLM-4.7 Flash | 25 | #529 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #6 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=greencode-effibench-correctness-default · How It Works · Data refreshed daily, snapshot 2026-10-27.