CESBench: leaderboard
Metric: Composite score (%; unweighted mean of the four task-type means: multiple-choice accuracy (209), judgment verdict-gated rubric score (67), scenario rubric score (63) and code task success on 572 hidden tests (41); 380 expert-written items on cryptographic engineering security for IoT devices; zero-shot, temperature 0, no tools, provider default reasoning mode; judgment and scenario answers scored by a Qwen3.5-397B judge from outside the evaluated set). Source: arxiv.org. Saturation forecast: Around December 2026. 11 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GLM-5.2 | 83.6 |
| 2 | Kimi K2.6 | 82.5 |
| 3 | Gemini 3.7 Flash | 81.4 |
| 4 | GPT-5.6 Luna | 80.8 |
| 5 | DeepSeek V4 Pro | 80.2 |
| 6 | MiniMax-M2.5 | 73.4 |
| 7 | DeepSeek V3 | 68.4 |
| 8 | Ling-flash-2.0 | 61.3 |
| 9 | GLM-4 32B (0414) | 60 |
| 10 | Llama 4 Maverick | 59.7 |
| 11 | Hunyuan A13B-Instruct | 54.4 |
Interactive version: theaggregate.ai/benchmark?slug=cesbench · How It Works · Data refreshed daily, snapshot 2026-09-26.