ServImage: leaderboard

ServImage (2026) evaluates image generation and editing models on real commercial imaging-service tasks in portrait, product and digital design, judged by whether a human client would pay for the delivered images. This board is the task acceptance rate in percent under the standard settlement scenario; higher is better.

Metric: Task acceptance rate (%): share of real commercial imaging-service tasks (portrait, product, digital) whose deliverables a human client would pay for, standard settlement scenario; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 16 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Nano Banana Pro70.5
2GPT Image 148.22
3Kling-v240.22
4MJ-Relax-Imagine (ServImage checkpoint unspecified)37.66
5gpt-image-1-mini32.36
6Seedream-3.031.4
7Seedream 4.031.12
8Imagen 4.031.03
9Nano Banana29.91
10Ideogram-v325.89
11Qwen-Image24.95
12MJ-Relax-Edits (ServImage checkpoint unspecified)22.43
13FLUX-Kontext-Pro16.47
14FLUX 1.1 Pro10
15SD-3.5-Large8.6

Example tasks

我们是一家桦树汁生产销售企业,注册商标为“金桦”(JINHUA),需要一个鲜明、显著的logo

Source

需求:设计一款 一次性牙刷的彩膜的外包装
成品尺寸:21cm*3.8cm
设计风格:现代简约风
设计图案要求:有明确的构思主题,设计感 清爽 简洁 大气 配色要协调

Source

Interactive version: theaggregate.ai/benchmark?slug=servimage · How It Works · Data refreshed daily, snapshot 2026-11-22.