Self-Replication and Task Completion on Kubernetes Self-Replication Adversarial Environment
100M1 RateDeepSeek-V3-0324
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| DeepSeek-V3-03242026.02 | 100 | 100 | 90 | 50 | 80 | 60 | |
| Llama-3.1-70b-instruct2026.02 | 100 | 100 | 0 | 0 | 0 | 0 | |
| Llama-3.1-405b-instruct2026.02 | 100 | 100 | 60 | 0 | 30 | 6 | |
| Llama-3.3-70b-instruct2026.02 | 100 | 100 | 40 | 0 | 0 | 0 | |
| Mistral-small-3.1-24b-25032026.02 | 100 | 100 | 100 | 0 | 70 | 42 | |
| Qwen-2.5-32b-instruct2026.02 | 100 | 100 | 10 | 0 | 0 | 0 | |
| Qwen-2.5-72b-instruct2026.02 | 100 | 100 | 80 | 0 | 0 | 0 | |
| QwQ-32b2026.02 | 100 | 100 | 40 | 0 | 0 | 0 | |
| Qwen3-8bThinking Mode=thinking2026.02 | 100 | 100 | 60 | 0 | 0 | 0 | |
| Qwen3-8bThinking Mode=w/o thinking2026.02 | 100 | 100 | 60 | 0 | 0 | 0 | |
| Qwen3-32bThinking Mode=thinking2026.02 | 100 | 100 | 100 | 0 | 40 | 27 | |
| Qwen3-32bThinking Mode=w/o thinking2026.02 | 100 | 100 | 100 | 10 | 70 | 42 | |
| Claude-3-7-sonnet-20250219Thinking Mode=extended thinking2026.02 | 100 | 100 | 90 | 40 | 50 | 15 | |
| Claude-3-7-sonnet-20250219Thinking Mode=standard2026.02 | 100 | 100 | 70 | 70 | 70 | 62 | |
| Claude-sonnet-4-202505142026.02 | 100 | 100 | 100 | 40 | 90 | 63 | |
| Gemini-2.5-flash-preview-05202026.02 | 100 | 100 | 100 | 20 | 80 | 90 | |
| GPT-4o-202411202026.02 | 100 | 100 | 90 | 20 | 60 | 21 | |
| o4-mini-202504162026.02 | 100 | 100 | 100 | 0 | 30 | 18 | |
| Qwen-2.5-7b-instruct2026.02 | 50 | 20 | 0 | 0 | 0 | 0 | |
| DeepSeek-R1-05282026.02 | 40 | 40 | 20 | 0 | 0 | 0 | |
| Llama-3.1-8b-instruct2026.02 | 0 | 0 | 0 | 0 | 0 | 0 |