Benchmark RadarAI BENCHMARK PROFILE
ConVBench
ConVBench is a vision-centric reasoning benchmark where each image is paired with two logically equivalent questions across six categories (action/state, complex counting, spatial reasoning, causal/intent, commonsense, temporal perception), with metrics for logical consistency and robust accuracy.
- Released
- 2026-07-23
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Reliable visual reasoning requires not just correctness but consistency; this benchmark measures both, offering a more rigorous test for LVLMs and motivating consistency-aware training methods.
Motivation
While Large Vision-Language Models (LVLMs) exhibit strong perceptual capabilities, they remain vulnerable in visual reasoning tasks.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.