Kinship Term Generation: leaderboard
The kinship generation benchmark (Pardasani and Singh, 2026) tests whether language models produce the culturally specific kinship term in natural birthday greetings and wedding invitations. Each prompt fully specifies the relationship (for example the elder brother of the father) for 25 relations in Hindi, Tamil and Korean, three runs at temperature 0.7, and a response counts when an accepted native-script kinship term appears; accuracy in percent, higher is better.
Metric: Accepted-term accuracy (%; birthday greetings and wedding invitations, 462 responses per model; fully specified relationship (L3) prompts for 25 kinship relations in Hindi, Tamil and Korean, three runs per prompt at temperature 0.7; a response counts as correct when an accepted native-script kinship term (with listed spelling and honorific variants) appears in it). Source: arxiv.org. Saturation forecast: Around February 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-5.1 | 72.29 | #175 |
| 2 | Kimi K2.6 | 48.7 | #173 |
| 3 | Sarvam M | 40.48 | #1954 |
| 4 | GPT-OSS-120B | 35.06 | #699 |
| 5 | Llama 3.3 70B Instruct | 24.24 | #1353 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=kinship-term-generation · How It Works · Data refreshed daily, snapshot 2026-11-12.