RoboRMBench - Lexical Substitution Mean Error: leaderboard
RoboRMBench mean error against the ground-truth progress score under lexical substitution. Lower is better. Source: arxiv.org/abs/2609.05401.
Metric: Mean Error. Source: arxiv.org. Saturation forecast: Around February 2027. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 0.68 | #103 |
| 2 | Claude Sonnet 4.6 | 1 | #94 |
| 3 | Gemini 2.5 Flash | 1.01 | #255 |
| 4 | Gemini 2.5 Flash Lite | 1.03 | #432 |
| 5 | GPT-5.1 | 1.04 | #145 |
| 6 | Gemma 3 12B | 1.08 | #674 |
| 7 | Claude Haiku 4.5 | 1.1 | #287 |
| 8 | Llama 4 Maverick | 1.11 | #486 |
| 9 | Llama 4 Scout | 1.18 | #648 |
| 10 | GPT-4o | 1.2 | #357 |
| 11 | Gemma 3 27B | 1.37 | #606 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=robormbench-lexical-substitution-mean-error · How It Works · Data refreshed daily, snapshot 2026-11-05.