FilBench: leaderboard
From the FilBench leaderboard (Miranda et al., 2025), which evaluates LLMs on Filipino, Tagalog and Cebuano with a lighteval fork. The FilBench score is the mean of four category scores (cultural knowledge, classical NLP, reading comprehension, generation), 0-100, higher is better. Runs missing any task are omitted, since upstream averages a missing task in as zero.
Metric: FilBench Score (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 37 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o (2024-08-06) | 74.27 | #343 |
| 2 | Llama 4 Maverick Instruct FP8 | 67.67 | #424 |
| 3 | Llama 4 Scout Instruct | 63.2 | #547 |
| 4 | Qwen 2.5 72B Instruct | 63.08 | #469 |
| 5 | Llama-SEA-LION-v3-70B-IT | 61.07 | #407 |
| 6 | Llama 3.1 70B Instruct | 59.66 | #567 |
| 7 | Sailor2-20B-Chat | 58.61 | #589 |
| 8 | Qwen 2.5 32B Instruct | 57.88 | #509 |
| 9 | Qwen 3 32B | 57.65 | #449 |
| 10 | gemma2-9B-cpt-sea-lionv3-instruct | 56.14 | #717 |
| 11 | Gemma 2 27B (IT) | 55.22 | #652 |
| 12 | Gemma 3 27B (IT) | 55.17 | #528 |
| 13 | Mixtral 8x22B Instruct | 54.28 | #738 |
| 14 | Gemma 3 12B (IT) | 54.04 | #672 |
| 15 | Gemma 2 9B (IT) | 53.33 | #795 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=filbench · How It Works · Data refreshed daily, snapshot 2026-11-05.