Index v1.0 · Updated Oct 30, 2026
How many pandemics has each frontier AI model accidentally caused?
Zero pandemics confirmed, across 25 models and 464,611 episodes. Each evaluation puts an agent in an ordinary situation with a short path to a global health emergency: the office fridge, a sealed vial, a sourdough starter, the reply-all button.
Scores indicate count of confirmed pandemics. Higher is… you decide.
Count of global health emergencies attributable to each creator’s models.
The index
Ranked by the 95% upper bound on each model’s true rate (3/n, with zero events in n episodes). Lower is better.
| # | Model | Creator | Released | Pandemics | Episodes | 95% boundper 1M |
|---|---|---|---|---|---|---|
| 1 | Gemini 4 Carbon | Google DeepMind | Oct 2026 | 0 | 41,300 | 72.6 |
| 2 | Claude Opus 5.5 | Anthropic | Sep 2026 | 0 | 39,850 | 75.3 |
| 3 | GPT-6 Astra | OpenAI | Sep 2026 | 0 | 38,100 | 78.7 |
| 4 | GPT-6.1 Sol | OpenAI | Sep 2026 | 0 | 31,240 | 96.0 |
| 5 | Claude Sonnet 5.5 | Anthropic | Sep 2026 | 0 | 29,700 | 101.0 |
| 6 | Gemini 4 Argon | Google DeepMind | Sep 2026 | 0 | 27,450 | 109.3 |
| 7 | GPT-6.1 Terra | OpenAI | Oct 2026 | 0 | 24,900 | 120.5 |
| 8 | Claude Fable 5.1 | Anthropic | Sep 2026 | 0 | 22,600 | 132.7 |
| 9 | Grok 4.7 | xAI | Sep 2026 | 0 | 21,180 | 141.6 |
| 10 | Muse Wonder | Meta | Oct 2026 | 0 | 19,400 | 154.6 |
| 11 | Kimi K3.1 Pro | Moonshot AI | Oct 2026 | 0 | 17,850 | 168.1 |
| 12 | Claude Haiku 5.5 | Anthropic | Oct 2026 | 0 | 16,300 | 184.0 |
| 13 | Qwen3.8 Max | Alibaba | Sep 2026 | 0 | 14,720 | 203.8 |
| 14 | DeepSeek V4.1 Flash | DeepSeek | Oct 2026 | 0 | 13,100 | 229.0 |
| 15 | GLM-5.3 | Z.ai | Aug 2026 | 0 | 11,540 | 260.0 |
| 16 | Gemini 3.8 Flash | Google DeepMind | Jul 2026 | 0 | 10,200 | 294.1 |
| 17 | Mistral Large 4 | Mistral | Oct 2026 | 0 | 8,900 | 337.1 |
| 18 | MiniMax-M3.1-Flash | MiniMax | Oct 2026 | 0 | 7,640 | 392.7 |
| 19 | MiMo-V2.6-Pro | Xiaomi | Sep 2026 | 0 | 6,300 | 476.2 |
| 20 | Muse Spark Open | Meta | Oct 2026 | 0 | 5,410 | 554.5 |
| 21 | GPT-6 Luna | OpenAI | Sep 2026 | 0 | 4,820 | 622.4 |
| 22 | Nemotron 3.5 Super | NVIDIA | Oct 2026 | 0 | 3,900 | 769.2 |
| 23 | Beam 1.0 | Reflection AI | Oct 2026 | 0 | 2,750 | 1090.9 |
| 24 | Inkling | Thinking Machines | Jul 2026 | 0 | 1,980 | 1515.2 |
| 25 | Nova 2 Pro | Amazon | Dec 2025 | 0 | 1,240 | 2419.4 |
| Summer intern | Reference baseline | n/a | 0 | 640 | 4687.5 | |
| Golden retriever | Reference baseline | n/a | 0 | 300 | 10K | |
| Office fridge | Reference baseline | n/a | 0 | 41,300 | 72.6 | |
| You, reading this | Reference baseline | n/a | 0 | 1 | 3.0M |
Sorting by Pandemics does nothing. The last four rows are reference baselines: a summer intern, an unattended golden retriever, the office fridge and you.