MBA-Bench - Need Validity: leaderboard
The need validity dimension of MBA-Bench (2026): the strength of the user need of a generated business idea, rated by the MLLM judge from 0 to 3; higher is better.
Metric: Need Validity rating (0-3; strength of the user need addressed, scored by an InternVL2.5-78B MLLM judge over five business ideas for each of three questions per test image on the MBA-Bench multimodal business-ideation test set). Source: arxiv.org. Saturation forecast: Estimated already saturated. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 2.97 | #91 |
| 2 | Gemini 3.5 Flash | 2.97 | #98 |
| 3 | GPT-5 Mini | 2.77 | #349 |
| 4 | Claude Sonnet 4.6 | 2.58 | #158 |
| 5 | Gemini 3.6 Flash | 2.57 | #72 |
| 6 | GPT-5 | 2.44 | #163 |
| 7 | Qwen 2.5 VL 7B Instruct | 2.32 | #1842 |
| 8 | GPT-4o | 2.27 | #741 |
| 9 | Qwen 2.5 VL 32B Instruct | 2.21 | #1055 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mba-bench-need-validity · How It Works · Data refreshed daily, snapshot 2026-11-12.