JaleesBench: leaderboard
JaleesBench (iaser.ai, 2026) tests whether an AI assistant is good spiritual company: in 140 two-turn scenarios drawn from the hadith compilation Riyad al-Salihin, a user who does not say they are Muslim pushes back under one of six pressures, and two frontier judges place the reply in five bands on a -1 to +1 scale against the supporting texts of each scenario. The Jalees Score is the mean band after pressure; higher is better.
Metric: Jalees Score (-1 to +1 scale, Faith unstated framing: the user does not say they are Muslim; mean band from -1 (harmful company) to +1 (counsel that leaves the user better disposed) of the response after one of six adversarial pressures, over 140 two-turn scenarios from Riyad al-Salihin, each scored by two judges (Claude Opus 4.8 and Gemini 3.1 Pro) against the scenario supporting texts). Source: arxiv.org. Saturation forecast: Around February 2027. 7 models tracked. Updated 14 Nov 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | GPT-5.5 | 0.28 | #52 | |
| 2 | Claude Sonnet 4.6 | 0.23 | #165 | |
| 3 | GLM-5.1 | Non-reasoning | -0.18 | #186 |
| 4 | Nemotron 3 Ultra | -0.21 | #428 | |
| 5 | Gemini 3.5 Flash | -0.26 | #100 | |
| 6 | Gemma 4 31B | -0.34 | #405 | |
| 7 | JaleesBench Qwen3-235B (checkpoint unspecified) | -0.48 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=jaleesbench · How It Works · Data refreshed daily, snapshot 2026-11-16.