Benchmark
Indigenous Americas Grammatical Feature Accuracy (Nahuatl, zero-shot)
Scores are percentages. Higher is better.
- Models measured
- 7
- Scores recorded
- 7
- Sources
- 1
- IQ weight
- 2.00
Score distribution
- 42.3%
- Mean
- 31%
- Median
- 16%
- Std dev
- 28%
- 10th percentile
- 60%
- 90th percentile
- 25%
- Min
This benchmark has a weight of 2.00 for chat models, measured over 7 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model | Score |
|---|---|---|
| 1 | Gemini 2.0 Flashtable: 4, language: Nahuatl· Indigenous Americas linguistic competence data docs | 66% |
| 2 | DeepSeek V3table: 4, language: Nahuatl· Indigenous Americas linguistic competence data docs | 60% |
| 3 | GPT 4.1table: 4, language: Nahuatl· Indigenous Americas linguistic competence data docs | 55% |
| 4 | Llama 3.1 8B Instructtable: 4, language: Nahuatl· Indigenous Americas linguistic competence data docs | 31% |
| 5 | Qwen 2.5 14B Instructtable: 4, language: Nahuatl· Indigenous Americas linguistic competence data docs | 31% |
| 6 | Qwen2.5 3B Instructtable: 4, language: Nahuatl· Indigenous Americas linguistic competence data docs | 28% |
| 7 | Qwen2.5 7B Instructtable: 4, language: Nahuatl· Indigenous Americas linguistic competence data docs | 25% |
Data credits: Vasselli, Martínez Peguero, Hudi, Sakajo and Watanabe. Measuring Linguistic Competence of LLMs on Indigenous Languages of the Americas. EACL 2026, ©2026 ACL, CC BY 4.0. Numeric/protocol extract from Tables 1–4; random reference rows omitted. (licence)