TRACK - Code Generation (Facts in Context): leaderboard
Knowledge-conflict reasoning in TRACK (Testing Reasoning Amid Conflicting Knowledge; Feng et al., EACL 2026): 500 BigCodeBench-derived coding problems that require specific library API versions, where the needed facts contradict or extend what the model learned in pre-training, answered with the model's own missing facts appended to the prompt. Holistic pass in percent (a correct final answer and reasoning that entails every required fact), higher is better.
Metric: Holistic pass (%; BigCodeBench-derived coding problems that require specific library API versions, 500 items; the final answer is correct and the reasoning entails every required fact; answered with the model's own missing facts appended to the prompt). Source: arxiv.org. Saturation forecast: Estimated already saturated. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4.1 Mini | 36.8 | #775 |
| 2 | Qwen 3 4B | 28.2 | #2297 |
| 3 | Qwen 3 8B | 27.8 | #1838 |
| 4 | Qwen 3 1.7B | 18.3 | #5551 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=track-code-generation-facts-in-context · How It Works · Data refreshed daily, snapshot 2026-11-12.