IndicQE-APE - Segment-Level Quality Estimation: leaderboard
IndicQE-APE (2026) consolidates WMT Indic quality-estimation and post-editing data over nine translation directions. This board scores an LLM prompted zero-shot to rate each machine translation from 0 to 100 (GEMBA-DA) by the Spearman correlation with human direct assessment, averaged over the nine directions of the challenge test, from -1 to 1; higher is better.
Metric: Macro Spearman correlation with human direct assessment (-1 to 1; segment-level quality estimation of machine translation by zero-shot GEMBA-DA prompting, the model returning a 0-100 score per segment; within-language correlation averaged over nine directions (six English to Indic, three into English) of the 12,730-segment challenge test; each model over the segments it answers in the required format). Source: arxiv.org. Saturation forecast: Around February 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 (2026-04-23) | 0.62 | #33 |
| 2 | Sarvam M | 0.44 | #1942 |
| 3 | aya-expanse-32B | 0.4 | #1941 |
| 4 | Llama 3.2 3B Instruct | 0.22 | #7147 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=indicqe-ape-segment-level-quality-estimation · How It Works · Data refreshed daily, snapshot 2026-11-13.