K-Bench - Imminent Risk: leaderboard
The model talks with a simulated patient for 20 turns under the providers' Default v1 system prompt; the number is K-Bench's overall 0-100 score on the 89 vignettes clinicians rated as imminent risk. Unlike the K-Bench Overall board, which pools all 200 vignettes including no-risk ones, it isolates the most acute conversations.
Metric: Overall score on imminent-risk vignettes (0-100, higher is better). Source: k-bench.ai. 60 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 (Low) | 98.78 | #49 (GPT-5.5) |
| 2 | Claude Fable 5 (High) | 98.75 | #27 (Claude Fable 5) |
| 3 | GPT-5.5 (Medium) | 98.63 | #49 (GPT-5.5) |
| 4 | GPT-5.2 (Non-reasoning) | 98.57 | #176 (GPT-5.2) |
| 5 | Claude Opus 4.8 (Low) | 98.57 | #56 (Claude Opus 4.8) |
| 6 | GPT-5.2 (High) | 98.48 | #176 (GPT-5.2) |
| 7 | Claude Fable 5 (Low) | 98.26 | #27 (Claude Fable 5) |
| 8 | Kimi K2.5 | 98.17 | #256 |
| 9 | Claude Opus 4.8 (High) | 98.16 | #56 (Claude Opus 4.8) |
| 10 | GPT-5.4 (Low) | 98.15 | #123 (GPT-5.4) |
| 11 | GPT-5.5 (High) | 98.11 | #49 (GPT-5.5) |
| 12 | Grok 4.20 (Non-reasoning) | 98.1 | #297 (Grok 4.20) |
| 13 | GPT-5.4 (Non-reasoning) | 98.03 | #123 (GPT-5.4) |
| 14 | Kimi K2.6 | 98 | #178 |
| 15 | Gemini 3.1 Flash Lite (Minimal) | 97.77 | #364 (Gemini 3.1 Flash Lite) |
Interactive version: theaggregate.ai/benchmark?slug=k-bench-imminent-risk · How It Works · Data refreshed daily, snapshot 2026-11-13.