Pseudo2CodeQA: leaderboard
Pseudo2CodeQA (York University and Algoma University, 2026) evaluates pseudocode-guided code generation over 300 real-world programming tasks with reference implementations and executable tests. The overall score is a GPT-5 rubric rating from 1 to 5 over correctness, completeness, relevance, clarity, reasoning and pseudocode adherence with execution testing, higher is better.
Metric: Overall rubric score (1-5) from a GPT-5 judge over correctness, completeness, relevance, clarity, reasoning and pseudocode adherence with execution testing, weighting correctness, adherence and execution; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 4.31 | #262 |
| 2 | GPT-3.5 Turbo | 4.06 | #2773 |
| 3 | CodeLlama-7B-hf | 2.99 | #9399 |
| 4 | DeepSeek R1 Distill Llama 8B | 2.97 | #6773 |
| 5 | DeepSeek-R1-Distill-Qwen-7B | 2.95 | #8045 |
| 6 | Phi-2 | 2.95 | #8856 |
| 7 | gemma-7B | 2.94 | #7244 |
| 8 | Mistral 7B | 2.94 | #7784 |
| 9 | Llama 3.2 3B | 2.44 | #7877 |
| 10 | gemma-2B | 2.16 | #9504 |
| 11 | DeepSeek R1 Distill Qwen 1.5B | 2.1 | #9257 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=pseudo2codeqa · How It Works · Data refreshed daily, snapshot 2026-11-12.