BanglaSocialBench - Address Terms: leaderboard
Metric: Accuracy (%) on the 982 Bangla address-term items: choosing the socially appropriate pronoun (correct if it matches the primary or a listed secondary acceptable form; instance-weighted random baseline 42.2) or nominal address form (four options), zero-shot Bangla prompts at temperature 0, culturally grounded scenarios written and verified by native Bangla speakers; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Flash | 81.16 | #237 |
| 2 | GPT-4o | 78.51 | #333 |
| 3 | Gemini 2.0 Flash | 76.17 | #331 |
| 4 | Claude Sonnet 4 | 75.46 | #194 |
| 5 | Gemma 3 27B | 73.42 | #596 |
| 6 | GPT-4o Mini | 65.99 | #588 |
| 7 | Qwen 2.5 72B Instruct | 64.77 | #436 |
| 8 | DeepSeek V3.1 | 64.77 | #260 |
| 9 | Llama 3.3 70B Instruct | 63.24 | #520 |
| 10 | Claude 3.5 Haiku | 61.91 | #553 |
| 11 | Gemma 3 12B | 54.99 | #666 |
| 12 | Llama 3 8B Instruct | 46.23 | #1115 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=banglasocialbench-address-terms · How It Works · Data refreshed daily, snapshot 2026-10-11.