AraGen - Conciseness: leaderboard
Conciseness dimension of AraGen's 3C3H evaluation of Arabic generative answers, judged by claude-sonnet-4 on the v3 question set: how free of unnecessary content an answer is, with answers judged incorrect scoring 0. Shown x100, higher is better.
Metric: 3C3H Conciseness (%). Source: huggingface.co. Saturation forecast: Around October 2027. 68 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O1 (2024-12-17) | 54.96 | #157 |
| 2 | O3 (2025-04-16) | 51.55 | #125 |
| 3 | Llama 3.3 70B Instruct | 46.76 | #541 |
| 4 | GPT-5 | 45.18 | #105 |
| 5 | O4 Mini (2025-04-16) | 44.38 | #183 |
| 6 | Llama 3.1 70B Instruct | 43.65 | #567 |
| 7 | O1 Mini (2024-09-12) | 43.6 | #412 |
| 8 | Llama 3.2 90B Vision Instruct | 40.97 | #621 |
| 9 | Claude 3.7 Sonnet (20250219) | 39.43 | #209 |
| 10 | O3 Mini (2025-01-31) | 36.43 | #249 |
| 11 | Claude Opus 4 (20250514) | 34.47 | #164 |
| 12 | Claude Sonnet 4.5 | 31.58 | #148 |
| 13 | Claude Sonnet 4 (20250514) | 31.53 | #223 |
| 14 | GPT-4o Mini (2024-07-18) | 30.5 | #568 |
| 15 | Falcon-H1-34B-Instruct | 29.24 | #658 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=aragen-conciseness · How It Works · Data refreshed daily, snapshot 2026-11-05.