MERA Original Paper - SimpleAr: leaderboard
SimpleAr from the original MERA paper's baseline runs (Fenogenova et al., ACL 2024, code v1.1.0): multi-digit addition problems answered 5-shot by greedy generation. Exact match in percent, higher is better.
Metric: Exact Match (%; 5-shot, greedy generation). Source: arxiv.org. Saturation forecast: Estimated already saturated. 19 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Yi 6B (Base) | 95.1 | #8358 |
| 2 | Mistral-7B-v0.1 | 95 | #6899 |
| 3 | davinci-002 | 92.7 | #7916 |
| 4 | Llama 2 13B Base | 91.1 | #7995 |
| 5 | Llama 2 7B Base | 83.9 | #8986 |
| 6 | ruGPT-3.5 | 2.9 | #9589 |
| 7 | mGPT-13B | 2.3 | #9566 |
| 8 | ruGPT-3-medium | 0.8 | #9690 |
| 9 | mGPT | 0.7 | #10733 |
| 10 | ruGPT3-large | 0.4 | #10217 |
| 11 | ruT5-base | 0 | #9694 |
| 12 | ruGPT-3-small | 0 | #9600 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mera-original-paper-simplear · How It Works · Data refreshed daily, snapshot 2026-11-12.