PaliBench - COMET: leaderboard
PaliBench (2026) asks models to translate 1,700 passages of the Pali Buddhist canon into English, scored against three independent human translations at once. This board reports COMET, a neural metric trained on human quality judgments, averaged over the three references and shown times 100; higher is better.
Metric: COMET (0-100): mean COMET score over the three references, times 100, 1,700 Pali canon passages translated into English, each scored against three human reference translations (Bodhi, Sujato, Thanissaro), corpus level, models through OpenRouter; higher is better. Source: arxiv.org. Saturation forecast: Around June 2027. 10 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash (Preview) | 73.1 | #152 |
| 2 | Gemini 3 Pro (Preview) | 72.9 | #119 |
| 3 | Claude Opus 4.5 | 72.4 | #160 |
| 4 | DeepSeek V3.2 | 71.8 | #419 |
| 5 | GPT-5.2 | 70.9 | #185 |
| 6 | Kimi K2.5 | 70.7 | #266 |
| 7 | Qwen 3 235B A22B 2507 Instruct | 70.6 | #616 |
| 8 | GLM-4.7 | 70.3 | #380 |
| 9 | Grok 4.1 Fast | 69.6 | #427 |
| 10 | Llama 3.3 70B Instruct | 68.1 | #1385 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=palibench-comet · How It Works · Data refreshed daily, snapshot 2026-11-21.