CRMArena: leaderboard
CRM task performance using the ReAct protocol across nine Salesforce workflows. The overall is the primary paper’s reported task mean; native and prompt-based function calling and Act are separate protocol boards with preserved checkpoint identities.
Metric: Mean task score (%). Source: huggingface.co. Source: arxiv.org. Saturation forecast: Estimated already saturated. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O1 (2024-12-17) | 57.7 | #149 |
| 2 | GPT-4o (2024-08-06) | 38.2 | #336 |
| 3 | DeepSeek R1 | 35.1 | #251 |
| 4 | Claude 3.5 Sonnet (20240620) | 34.3 | #299 |
| 5 | Llama 3.1 405B Instruct | 33.8 | #457 |
| 6 | GPT-4o Mini | 28.3 | #599 |
| 7 | Llama 3.1 70B Instruct | 27.8 | #561 |
| 8 | Claude 3 Sonnet (20240229) | 17.3 | #677 |
| 9 | Llama 3.1 8B Instruct | 3.1 | #1035 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=crmarena · How It Works · Data refreshed daily, snapshot 2026-10-20.