G-IdiomAlign - Gloss-Contrastive Generation (With Gloss): leaderboard

Metric: Acc@0.80 (%; the model writes one target-language idiom for a source idiom, correct when its Qwen3-Embedding-8B cosine similarity to the reference idiom is at least 0.80, greedy decoding, macro-average over 72 translation directions; source idiom plus its English gloss). Source: arxiv.org. Saturation forecast: Around 2032. 7 models tracked.

Top models

#ModelScore
1DeepSeek V3.2 (Thinking)29.63
2Gemini 2.5 Pro27.11
3Claude Haiku 4.525.66
4DeepSeek V3.2 (Non-reasoning)23.57
5Ministral-8B-Instruct-241021.22
6Qwen 3 8B (Thinking)20.86
7Qwen 3 8B (Non-reasoning)18.8

Interactive version: theaggregate.ai/benchmark?slug=g-idiomalign-gloss-contrastive-generation-with-gloss · How It Works · Data refreshed daily, snapshot 2026-09-26.