BhashaBench - Krishi: leaderboard
From BharatGen's BhashaBench V1 leaderboard, which scores models on English and Hindi questions sourced from Indian government and domain-specific exams. BhashaBench-Krishi tests Indian agriculture knowledge; this board is accuracy over all questions in both languages (%), higher is better. Flagged runs are skipped.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Around February 2027. 31 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 73.5 | #357 |
| 2 | Qwen 3 235B A22B 2507 Instruct | 72.7 | #315 |
| 3 | GPT-OSS-120B | 69.41 | #346 |
| 4 | DeepSeek V3.1 | 59.73 | #275 |
| 5 | Gemma 2 27B | 58.14 | #803 |
| 6 | Gemma 2 9B | 53.2 | #924 |
| 7 | Gemma 2 27B (IT) | 52.5 | #652 |
| 8 | Gemma 2 9B (IT) | 46.69 | #795 |
| 9 | Llama 3.1 8B Instruct | 44.98 | #1017 |
| 10 | Pangea-7B | 44.93 | #1223 |
| 11 | GPT-OSS-20B | 44.73 | #523 |
| 12 | Qwen 2.5 3B | 42.45 | #1177 |
| 13 | Indic-gemma-7B-finetuned-sft-Navarasa-2.0 | 40.73 | #1372 |
| 14 | Qwen 2.5 3B Instruct | 39.9 | #1159 |
| 15 | Gemma 2 2B | 38.78 | #1466 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=bhashabench-krishi · How It Works · Data refreshed daily, snapshot 2026-11-05.