BALSAM - Information Extraction: leaderboard
Metric: Overall score (0-100, LLM-judged generation and multiple choice). Source: benchmarks.ksaa.gov.sa. 29 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemma 4 31B (IT) | 57.49 |
| 2 | GLM-5 (Thinking) | 53.79 |
| 3 | GLM-4.7 (Reasoning) | 53.15 |
| 4 | Grok 4.1 Fast (Reasoning) | 52.84 |
| 5 | GPT-5.2 | 51.1 |
| 6 | Llama 4 Maverick Instruct | 50.66 |
| 7 | Kimi K2 Instruct (0905) | 50.36 |
| 8 | DeepSeek V3.2 | 50.07 |
| 9 | Command A 03 2025 | 49.92 |
| 10 | Fanar-C-2-27B | 49.34 |
| 11 | Qwen 3 235B A22B | 49.32 |
| 12 | AceGPT-v2-8B-Chat | 49.27 |
| 13 | Llama 3.3 70B Instruct | 49.17 |
| 14 | DeepSeek V4 Pro (Reasoning) | 47.59 |
| 15 | SILMA-9B-Instruct-v1.0 | 46.9 |
Interactive version: theaggregate.ai/benchmark?slug=balsam-information-extraction · How It Works · Data refreshed daily, snapshot 2026-09-19.