ITBench-AA: leaderboard
Artificial Analysis implementation of IBM's ITBench SRE benchmark, testing AI agents on Kubernetes incident root-cause analysis from offline incident snapshots.
Metric: Average Precision at Full Recall (self-reported). Source: benchmarklist.com. Status: saturation imminent. 23 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama 3.3 70B Instruct | 60 |
| 2 | Claude Opus 4.7 (Adaptive Reasoning, Max Effort) | 46.7 |
| 3 | GPT-5.5 (xHigh) | 45.8 |
| 4 | Qwen 3.7 Max | 42.5 |
| 5 | Gemini 3.5 Flash (High) | 40.3 |
| 6 | GLM-5.1 (Thinking) | 40.3 |
| 7 | Claude Sonnet 4.6 (Adaptive Reasoning, Max Effort) | 39.8 |
| 8 | DeepSeek V4 Pro (Reasoning, Max Effort) | 38.3 |
| 9 | MiMo-V2.5-Pro | 38.2 |
| 10 | Gemma 4 31B (Reasoning) | 37.3 |
| 11 | GPT-5.4 Mini (xHigh) | 35.2 |
| 12 | GPT-5.4 (xHigh) | 34.5 |
| 13 | Qwen 3.5 397B A17B (Reasoning) | 34.1 |
| 14 | Grok 4.3 (High) | 32.7 |
| 15 | DeepSeek V4 Flash (Reasoning, Max Effort) | 31.5 |
Interactive version: theaggregate.ai/benchmark?slug=itbench-aa · How It Works · Data refreshed daily, snapshot 2026-09-05.