ESG Benchmark: leaderboard
Expert-built ESG question answering from the ESGAgent paper (Zhao et al., 2026), grounded in the sustainability and financial reports of Dow Jones constituents: 132 Level 1 questions from CFA ESG-certificate practice material and 114 Level 2 compositional tasks with PDF, CSV, audio and video attachments. Scored by exact-answer accuracy (%) over all 246 questions; higher is better. The paper does not say how the baseline LLMs received the Level 2 attachments.
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Around February 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 80.89 | #159 |
| 2 | Gemini 3 Pro | 78.05 | #135 |
| 3 | GPT-5 | 77.24 | #163 |
| 4 | DeepSeek R1 | 77.24 | #467 |
| 5 | O3 | 76.83 | #224 |
| 6 | Gemini 2.5 Pro | 76.42 | #262 |
| 7 | Claude Opus 4.5 | 76.02 | #152 |
| 8 | GPT-4.1 | 74.39 | #492 |
| 9 | GPT-4o | 73.17 | #741 |
| 10 | GPT-5.2 | 72.76 | #172 |
| 11 | DeepSeek V3 | 72.76 | #705 |
| 12 | Gemini 2.5 Flash | 72.36 | #452 |
| 13 | GPT-5.1 | 70.33 | #225 |
| 14 | Claude Sonnet 4.5 | 69.92 | #244 |
| 15 | Claude Haiku 4.5 | 69.11 | #568 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=esg-benchmark · How It Works · Data refreshed daily, snapshot 2026-11-12.