McBE
Harbor: pendingAs large language models (LLMs) are increasingly applied to various NLP tasks, their inherent biases are gradually disclosed.
- Samples
- N/A
- Released
- 2025
- Scores
- 0
APRiL
An Open-ended Benchmark for Chinese Social Awareness in Large Language Models
APRiL evaluates Chinese social awareness in LLMs using 2,196 open-ended instances across six categories, assessing scenario analysis and response generation.
How to run
ltb · april
$ npm i -g lettokenburn-cli$ ltb bench info april
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
9 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| DS-V4-Pro | 83.6 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| GPT-5.1 | 82.6 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| Ministral-14B | 79.2 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| Gemini 3 Pro | 77.0 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| Ministral 8B | 76.7 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| Kimi K2.5 | 74.9 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| Gemini 3.8 Flash | 74.5 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| DS-V4-Flash | 72.4 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| Ministral 3B | 56.0 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsLoading community runs…
About
The benchmark comprises 2,196 Social Awareness Evaluation Instances covering implicit intention, power relations, norms, conflict, fairness, and support. It features two tasks: Social Scenario Analysis and Socially Aware Response.
Scoring uses instance-specific rubrics to evaluate understanding and response appropriateness. Experiments on 12 models reveal a gap between scenario understanding and response quality, with better analysis improving responses.