KrishokChat - Treatment QA: leaderboard
Metric: Treatment QA correct rate (%): the recommended treatment matches the provenance-tracked chemical, dosage and unit exactly (correct, omission or hallucination verdict against the chemical trace), 346 test questions in standard Bengali and five regional dialects across fertilizer, disease, pest, IPM, seed, food-safety and herbicide domains, zero-shot with a standardized Bengali agricultural system prompt, no few-shot examples, closed-book (question only); higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 5 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemma 4 26B A4B (IT) | 38.73 |
| 2 | GPT-OSS-120B | 32.92 |
| 3 | Qwen 2.5 7B Instruct | 13.29 |
| 4 | Llama 3.1 8B Instruct | 12.43 |
Interactive version: theaggregate.ai/benchmark?slug=krishokchat-treatment-qa · How It Works · Data refreshed daily, snapshot 2026-09-29.