SAFARI - Task 1 (Few-Shot): leaderboard
Automotive hazard analysis and risk assessment (ISO 26262 HARA) with few-shot prompting: the generated hazardous event, hazard and persons at risk scored on Hazard Logic Mapping, Physical Consequence Consistency and Risk Coverage Completeness (mean of the three), judged by Qwen3.6-Max calibrated against certified functional-safety engineers. Higher is better. Source: arxiv.org/abs/2609.20584.
Metric: Judge score (%, mean of HLM/PCC/RCC, judge Qwen3.6-Max). Source: arxiv.org. Saturation forecast: Around May 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-5.1 | 72.6 | #109 |
| 2 | Gemini 3.5 Flash | 71.5 | #66 |
| 3 | DeepSeek V4 Flash | 71.4 | #155 |
| 4 | Qwen 3.5 397B A17B | 69.1 | #154 |
| 5 | MiniMax-M2.7 | 69 | #270 |
| 6 | DeepSeek V4 Pro | 68.5 | #112 |
| 7 | Qwen 3.5 122B A10B | 65.6 | #184 |
| 8 | GPT-5.5 | 65.3 | #33 |
| 9 | GPT-5.4 Mini | 59.8 | #228 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=safari-task-1-few-shot · How It Works · Data refreshed daily, snapshot 2026-11-05.