Persian Fluency Bench - Context Retention: leaderboard
Metric: Judge Score (%, judge google/gemini-2.5-flash). Source: github.com. 16 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemma 3 12B (IT) | 85.9 |
| 2 | Qwen 3 30B A3B 2507 Instruct | 78.5 |
| 3 | Gemini 2.5 Flash | 76.7 |
| 4 | Qwen 3 Next 80B A3B Instruct | 76.6 |
| 5 | Gemma 3 4B (IT) | 76.5 |
| 6 | Gemma 3 27B (IT) | 73.5 |
| 7 | Gemma 3n E4B (IT) | 73.1 |
| 8 | ERNIE 4.5 21B A3B | 59.4 |
| 9 | granite-4.0-h-micro | 32.3 |
| 10 | Qwen 3 4B | 17.1 |
| 11 | Llama 3.2 3B Instruct | 16.6 |
| 12 | Mixtral 8x7B Instruct | 15 |
| 13 | Ministral 3B | 13.9 |
| 14 | LFM2 8B A1B | 3.9 |
Interactive version: theaggregate.ai/benchmark?slug=persian-fluency-bench-context-retention · How It Works · Data refreshed daily, snapshot 2026-09-20.