G-IdiomAlign - Gloss-Contrastive Generation (No Gloss): leaderboard
G-IdiomAlign (Ye et al., 2026): Given only a source idiom, the model writes one meaning-equivalent idiom in the target language. A prediction counts as correct when its Qwen3-Embedding-8B cosine similarity to the reference idiom is at least 0.80, a semantic proxy that can miss valid synonymous idioms; accuracy in percent (higher is better), macro-averaged over 72 translation directions.
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 only). Source: arxiv.org. Saturation forecast: Around 2029. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V3.2 (Thinking) | 24.66 | #403 (DeepSeek V3.2) |
| 2 | Claude Haiku 4.5 | 20.47 | #568 |
| 3 | DeepSeek V3.2 (Non-reasoning) | 19.9 | #403 (DeepSeek V3.2) |
| 4 | Gemini 2.5 Pro | 19.7 | #262 |
| 5 | Qwen 3 8B (Thinking) | 14.04 | #1838 (Qwen 3 8B) |
| 6 | Ministral-8B-Instruct-2410 | 12.96 | #4720 |
| 7 | Qwen 3 8B (Non-reasoning) | 11.59 | #1838 (Qwen 3 8B) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=g-idiomalign-gloss-contrastive-generation-no-gloss · How It Works · Data refreshed daily, snapshot 2026-11-12.