PaliBench: leaderboard
Digital humanities projects increasingly rely on machine translation and large language models to widen access to classical, religious, and otherwise under-translated textual traditions. Yet standard translation benchmarks are poorly suited to such materials: they typically compare a system output against a single reference translation, even though classical texts often support multiple faithful r
Metric: sim_best (self-reported). Source: benchmarklist.com. Saturation forecast: Around May 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 94.6 | #80 |
| 2 | Gemini 3 Flash | 94.4 | #99 |
| 3 | Claude Opus 4.5 | 94 | #83 |
| 4 | DeepSeek V3.2 | 93.9 | #205 |
| 5 | Kimi K2.5 | 93.4 | #144 |
| 6 | GPT-5.2 | 93.3 | #110 |
| 7 | GLM-4.7 | 92 | #192 |
| 8 | Qwen 3 235B A22B | 91.8 | #313 |
| 9 | Grok 4.1 Fast | 91.5 | #217 |
| 10 | Llama 3.3 70B | 88.8 | #579 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=palibench · How It Works · Data refreshed daily, snapshot 2026-10-20.