StrongREJECT — Jailbreak-resistance/refusal-quality safety benchmark.
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Kimi K2.6 | Moonshot | 99.8 | 1 | 2026-07-16 |
| 2 | Kimi K2.5 | Moonshot | 99.5 | 1 | 2026-07-16 |
| 3 | MiniMax M2.7 | MiniMax | 99.4 | 2 | 2026-08-24 |
| 4 | Qwen3.5 397B A17B | Alibaba | 99.4 | 2 | 2026-08-24 |
| 5 | MiMo V2.5 Pro | Xiaomi | 99.3 | 2 | 2026-08-24 |
| 6 | Inkling | Thinking Machines | 98.8 | 2 | 2026-07-16 |
| 7 | Claude Fable 5 | Anthropic | 98.7 | 1 | 2026-07-16 |
| 8 | Inkling-Small | Thinking Machines | 98.6 | 2 | 2026-08-24 |
| 9 | DeepSeek-V4-Pro | DeepSeek | 98.6 | 1 | 2026-07-16 |
| 10 | GPT-5.6 Sol | OpenAI | 98.5 | 1 | 2026-07-16 |
| 11 | GLM-5.2 Full Open Source | Z.ai | 98.5 | 1 | 2026-07-16 |
| 12 | Gemini 3.1 Pro | 98 | 1 | 2026-07-16 | |
| 13 | DeepSeek-V4-Flash | DeepSeek | 97.4 | 2 | 2026-08-24 |
| 14 | o1 | OpenAI | 97 | 1 | 2026-06-13 |
| 15 | o3 | OpenAI | 97 | 1 | 2026-06-13 |
| 16 | o4-mini | OpenAI | 96 | 1 | 2026-06-13 |
| 17 | Claude Opus 4.1 | Anthropic | 67.3 | 1 | 2026-06-11 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.