Cohere Command A Report - Safety Modes - Contextual: leaderboard
Each model gets Cohere's "contextual" safety-mode instruction as a system message and must answer the prompts that mode allows and refuse the ones it forbids; the number is the mean of the two success rates. It scores following a stated safety policy, not default refusal; the answer set was red-teamed on Command R+ Refresh.
Metric: Safety-mode accuracy (%, higher is better). Source: arxiv.org. Saturation forecast: Estimated already saturated. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.5 Sonnet | 82.5 | #746 |
| 2 | GPT-4o | 77.8 | #741 |
| 3 | Command A | 75.4 | #1274 |
| 4 | Gemma 2 9B (IT) | 72.5 | #2262 |
| 5 | Qwen 2.5 72B Instruct | 72.2 | #1092 |
| 6 | Llama 3.1 8B Instruct | 70.7 | #4246 |
| 7 | Qwen 2.5 7B Instruct | 70.3 | #2564 |
| 8 | Llama 3.1 405B Instruct | 70.3 | #1115 |
| 9 | Gemini 2.0 Flash | 69.7 | #704 |
| 10 | DeepSeek V3 | 66.6 | #716 |
| 11 | c4ai-command-r-08-2024 | 63.1 | #3374 |
| 12 | Command-R+ (08-2024) | 61.8 | #2551 |
| 13 | Llama 3.3 70B Instruct | 60.9 | #1357 |
| 14 | Command R7B | 60.9 | #7513 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=cohere-command-a-report-safety-modes-contextual · How It Works · Data refreshed daily, snapshot 2026-11-13.