BALSAM - Overall: leaderboard
BALSAM Arabic LLM benchmark by the Saudi Data and AI Authority and the King Salman Global Academy for Arabic Language (dataset version 2): mean of the 14 category overall scores, which blend LLM-judged generated answers with multiple-choice answers. 0-100, higher is better; models missing a category are left out.
Metric: Mean of category overall scores (0-100). Source: benchmarks.ksaa.gov.sa. Saturation forecast: Around 2031. 30 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 59.61 | #115 |
| 2 | GLM-5 (Thinking) | 59.29 | #149 (GLM-5) |
| 3 | GLM-4.7 (Reasoning) | 58.01 | #202 (GLM-4.7) |
| 4 | GPT-OSS-120B (Reasoning) | 55.91 | #346 (GPT-OSS-120B) |
| 5 | AceGPT-v2-8B-Chat | 55.9 | #746 |
| 6 | DeepSeek V3.2 | 55.33 | #210 |
| 7 | Command A 03 2025 | 55.16 | #530 |
| 8 | Grok 4.1 Fast (Reasoning) | 55.03 | #227 (Grok 4.1 Fast) |
| 9 | Nova Pro (v1) | 54.76 | #508 |
| 10 | Kimi K2 Instruct (0905) | 54.57 | #267 |
| 11 | Qwen 3 235B A22B | 53.84 | #323 |
| 12 | Fanar-C-2-27B | 53.17 | #637 |
| 13 | Llama 4 Maverick Instruct | 52.92 | #467 |
| 14 | Llama 3.3 70B Instruct | 52.25 | #541 |
| 15 | Claude Opus 4.5 | 50.62 | #87 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=balsam-overall · How It Works · Data refreshed daily, snapshot 2026-11-05.