LLaDA1.5 Release - Arena-Hard: leaderboard
LLaDA1.5 release evaluation, Table4 per-model block/answer lengths retained in source; LLaDA8BInstruct is the SFT predecessor, final1.5 is SFT+VRPO; no ablation checkpoint merge. Metric: Benchmark score (%). Higher is better. Source: https://arxiv.org/html/2505.19223v1
Metric: Benchmark score (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 2 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | LLaDA 1.5 | 14.3 | |
| 2 | LLaDA-8B-Instruct | 10 |
Interactive version: theaggregate.ai/benchmark?slug=llada1-5-release-arena-hard · How It Works · Data refreshed daily, snapshot 2026-10-17.