CodeAssay: leaderboard

Metric: Final hidden-test correctness (%; share of 185 audited Python tasks in ten software-engineering categories whose scored solution passes every hidden test; one generation from the description, signature and a public-test excerpt plus at most one repair guided by public-test failures; provider-default reasoning, temperature 0 where accepted, 16,384-token output limit; standard system prompt). Source: arxiv.org. Saturation forecast: Estimated already saturated. 7 models tracked.

Top models

#ModelScore
1GPT-5.6 Sol98.9
2Claude Sonnet 598.9
3Claude Haiku 4.596.8
4Claude Sonnet 4.595.1
5Gemini 2.5 Flash91.4
6GPT-4o87
7GPT-4o Mini77.3

Interactive version: theaggregate.ai/benchmark?slug=codeassay · How It Works · Data refreshed daily, snapshot 2026-09-29.