SAKE - Zero-Shot - Behavioral Patterns: leaderboard
Evaluates software architectural knowledge through multiple-choice questions on the paper's 2,114 scored test items, with temperature zero and short answer limits. The paper reports model display labels without exact endpoint snapshots; those labels are retained. Each task series keeps its prompting condition. The five-shot examples are held out from both scored populations; task results share the parent suite. This series reports SAKE - Zero-Shot - Behavioral Patterns using Accuracy (%).
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 95.3 | #67 |
| 2 | Claude Sonnet 4.6 | 94.9 | #94 |
| 3 | GPT-5.4 | 94.9 | #83 |
| 4 | Qwen 3 235B A22B | 94.9 | #323 |
| 5 | Gemini 3.1 Flash Lite | 94.6 | #200 |
| 6 | DeepSeek V3.2 | 93.1 | #210 |
| 7 | Grok 4.20 | 93.1 | #179 |
| 8 | Grok 4.1 Fast | 92.7 | #227 |
| 9 | Mistral Small 4 | 92.4 | #522 |
| 10 | Claude Haiku 4.5 | 91.6 | #287 |
| 11 | GPT-5.4 Nano | 91.3 | #359 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=sake-zero-shot-behavioral-patterns · How It Works · Data refreshed daily, snapshot 2026-11-05.