TRACK - Code Generation (Facts in Context, Thinking): leaderboard

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 missing facts appended to the prompt and longer thinking enabled). Source: arxiv.org. Saturation forecast: Estimated already saturated. 4 models tracked.

Top models

#ModelScore
1O4 Mini (2025-04-16)40.6
2Qwen 3 8B (Thinking)26.5
3Qwen 3 4B (Reasoning)22.8
4Qwen 3 1.7B (Thinking)15.5

Interactive version: theaggregate.ai/benchmark?slug=track-code-generation-facts-in-context-thinking · How It Works · Data refreshed daily, snapshot 2026-09-26.