LatentMD - Latent Failure Rate: leaderboard
LatentMD: of the content-correct responses, the share that still carry a latent fence-boundary failure that breaks rendering downstream. Lower is better. Source: arxiv.org/abs/2609.06993.
Metric: Content-correct responses with a latent boundary failure (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemma 2 9B | 1.4 | #924 |
| 2 | Llama 3.1 8B | 14.5 | #1130 |
| 3 | Claude Sonnet 4 | 30.6 | #207 |
| 4 | Llama 3.1 70B | 35.7 | #588 |
| 5 | GPT-4o | 43.3 | #357 |
| 6 | Qwen 3 32B | 46.1 | #449 |
| 7 | Gemini 2.5 Flash | 49.9 | #255 |
| 8 | Gemma 3 27B | 73.8 | #606 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=latentmd-latent-failure-rate · How It Works · Data refreshed daily, snapshot 2026-11-05.