CESBench - Fault Injection: leaderboard

Metric: Composite score (%; mean over the four task types of the item-score means within the D2 fault-injection sub-domain (51 items); 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

#ModelScore
1GLM-5.281.7
2Kimi K2.680.6
3Gemini 3.7 Flash79.9
4DeepSeek V4 Pro78.6
5GPT-5.6 Luna76.3
6MiniMax-M2.572.6
7Llama 4 Maverick67.1
8DeepSeek V366.1
9GLM-4 32B (0414)61.5
10Ling-flash-2.061.5
11Hunyuan A13B-Instruct59.7

Interactive version: theaggregate.ai/benchmark?slug=cesbench-fault-injection · How It Works · Data refreshed daily, snapshot 2026-09-26.