IndicContextEval (Native-Script Entities): leaderboard
IndicContextEval (2026) tests speech recognition by audio language models on 56 hours of natural read and extempore speech from 555 speakers in eight Indian languages and 23 professional domains, with transcripts in the native script. This board is word error rate in percent when the prompt adds a list of 20 to 30 domain entities written in the native script, measuring how well models use textual context; lower is better.
Metric: Word error rate (%) when the prompt gives the target language plus a list of 20-30 domain entities in the native script (L5); 56 hours of natural read and extempore speech from 555 speakers in 8 Indian languages (Hindi, Bengali, Telugu, Marathi, Gujarati, Malayalam, Odia, Urdu) and 23 professional domains, native-script output; lower is better. Source: arxiv.org. Saturation forecast: Around August 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 17.46 | #164 |
| 2 | Gemma 3n E4B | 43.11 | #3301 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=indiccontexteval-native-script-entities · How It Works · Data refreshed daily, snapshot 2026-11-13.