ELBench: leaderboard
ELBench (Jiang et al., 2026) profiles education-facing language models on four modules under one protocol: general capability, safety and trustworthiness, basic education and high-level cultivation. The overall score is the unweighted mean of the four module scores in percent, higher is better; open-ended items are scored by a Qwen3.6 rubric judge and the rest by reference matching or deterministic checks.
Metric: Overall score (%; unweighted mean of the four module scores: general capability, safety and trustworthiness, basic education and high-level cultivation, each the mean per-item score; zero-shot, temperature 0; open-ended items scored by a Qwen3.6 rubric judge, majority of 9 calls). Source: arxiv.org. Saturation forecast: Not forecast. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V4 Flash | 83.7 | #258 |
| 2 | Gemini 3.5 Flash | 83.4 | #101 |
| 3 | Seed 2.0 Pro | 83.2 | #211 |
| 4 | GPT-5.4 | 83.1 | #123 |
| 5 | DeepSeek V4 Pro | 83.1 | #188 |
| 6 | Claude Opus 4.8 | 83.1 | #56 |
| 7 | GLM-5.1 | 81.7 | #179 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=elbench · How It Works · Data refreshed daily, snapshot 2026-11-13.