VakQA (English, Seamless MT): leaderboard
VākQA (Akkiraju et al., 2026) is a Telugu spoken factoid question-answering benchmark of 2,001 quiz-style questions across six domains with human-verified reference answers. Each answer is scored from 1 to 5 for correctness against the reference by Gemini-2.5-Flash as judge, and the board is the mean score; higher is better. Input: the Telugu transcription machine-translated into English by Seamless MT.
Metric: Judge score (1-5) (2,001 Telugu spoken factoid questions from quiz-style interactions across six domains, with human-verified reference answers; each answer scored 1 to 5 for correctness against the reference by Gemini-2.5-Flash as judge (the best-aligned judge in the human study); mean over the questions; input: the human-verified Telugu transcription machine-translated into English by Seamless MT). Source: arxiv.org. Saturation forecast: Estimated already saturated. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Flash | 2.74 | #452 |
| 2 | Gemma 3 27B | 2.45 | #1638 |
| 3 | Sarvam M | 2.29 | #1954 |
| 4 | Gemma 3 12B | 2.28 | #2013 |
| 5 | Gemma 3 4B | 1.84 | #5104 |
| 6 | Qwen 3 4B | 1.74 | #2297 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=vakqa-english-seamless-mt · How It Works · Data refreshed daily, snapshot 2026-11-12.