BALSAM - Information Extraction: leaderboard
BALSAM Arabic benchmark (SDAIA/KSAA, dataset version 2), Information Extraction category: analysing texts to extract specific, structured information. The site's overall score blends LLM-judged generated answers with multiple-choice answers across the category's tasks; 0-100, higher is better.
Metric: Overall score (0-100, LLM-judged generation and multiple choice). Source: benchmarks.ksaa.gov.sa. Saturation forecast: Around 2032. 36 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-5 (Thinking) | 53.79 | #149 (GLM-5) |
| 2 | GLM-4.7 (Reasoning) | 53.15 | #202 (GLM-4.7) |
| 3 | Grok 4.1 Fast (Reasoning) | 52.84 | #227 (Grok 4.1 Fast) |
| 4 | GPT-5.2 | 51.1 | #115 |
| 5 | Llama 4 Maverick Instruct | 50.66 | #467 |
| 6 | Kimi K2 Instruct (0905) | 50.36 | #267 |
| 7 | DeepSeek V3.2 | 50.07 | #210 |
| 8 | Command A 03 2025 | 49.92 | #530 |
| 9 | Fanar-C-2-27B | 49.34 | #637 |
| 10 | Qwen 3 235B A22B | 49.32 | #323 |
| 11 | AceGPT-v2-8B-Chat | 49.27 | #746 |
| 12 | Llama 3.3 70B Instruct | 49.17 | #541 |
| 13 | GPT-OSS-120B (Reasoning) | 48.6 | #346 (GPT-OSS-120B) |
| 14 | Nova Pro (v1) | 48.29 | #508 |
| 15 | SILMA-9B-Instruct-v1.0 | 46.9 | #927 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=balsam-information-extraction · How It Works · Data refreshed daily, snapshot 2026-11-05.