AlmanBench - Acceptance: leaderboard
AlmanBench is a Standard German to Alman translation benchmark: 1,029 public items, each with a licensed set of acceptable Alman renderings. Acceptance is a normalised exact match of the model’s output against that set, which is the benchmark’s primary score.
Metric: Acceptance Rate (%). Source: huggingface.co. Saturation forecast: Around January 2027. 32 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 (xHigh) | 95.43 | #31 (GPT-5.5) |
| 2 | GPT-5.6 Sol (Max) | 95.14 | #19 (GPT-5.6 Sol) |
| 3 | Claude Opus 5 (Max) | 94.95 | #12 (Claude Opus 5) |
| 4 | Claude Fable 5 (High) | 93.97 | #16 (Claude Fable 5) |
| 5 | Claude Fable 5 (Max) | 93.88 | #16 (Claude Fable 5) |
| 6 | GPT-5.6 Sol (xHigh) | 93.49 | #19 (GPT-5.6 Sol) |
| 7 | DeepSeek V4 Flash (0731) (Max) | 93 | #78 (DeepSeek V4 Flash (0731)) |
| 8 | Kimi K3 (Max) | 90.96 | #30 (Kimi K3) |
| 9 | GLM-5.2 | 90.77 | #73 |
| 10 | DeepSeek V4.1 Flash (Max) | 90.48 | #36 (DeepSeek V4.1 Flash) |
| 11 | Inkling (Max) | 89.41 | #126 (Inkling) |
| 12 | GPT-5.6 Luna (xHigh) | 87.76 | #68 (GPT-5.6 Luna) |
| 13 | GPT-5.6 Terra (xHigh) | 87.66 | #41 (GPT-5.6 Terra) |
| 14 | Kimi K2.7 Code | 86.98 | #89 |
| 15 | DeepSeek V4 Pro | 85.33 | #102 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=almanbench-acceptance · How It Works · Data refreshed daily, snapshot 2026-10-20.