Vibe-Eval: A hard evaluation suite for measuring progress of multimodal language models

TL;DR

Vibe-Eval is a comprehensive benchmark designed to rigorously evaluate multimodal chat models through challenging prompts, revealing their limitations and providing tools for ongoing assessment.

Contribution

This paper introduces Vibe-Eval, a novel open benchmark with hard prompts and expert responses to evaluate and rank the capabilities of multimodal language models.

Findings

Over 50% of hard prompts are answered incorrectly by frontier models

Automatic evaluation correlates roughly with human judgment

Vibe-Eval provides a challenging and open-ended assessment framework

Abstract

We introduce Vibe-Eval: a new open benchmark and framework for evaluating multimodal chat models. Vibe-Eval consists of 269 visual understanding prompts, including 100 of hard difficulty, complete with gold-standard responses authored by experts. Vibe-Eval is open-ended and challenging with dual objectives: (i) vibe checking multimodal chat models for day-to-day tasks and (ii) rigorously testing and probing the capabilities of present frontier models. Notably, our hard set contains >50% questions that all frontier models answer incorrectly. We explore the nuances of designing, evaluating, and ranking models on ultra challenging prompts. We also discuss trade-offs between human and automatic evaluation, and show that automatic model evaluation using Reka Core roughly correlates to human judgment. We offer free API access for the purpose of lightweight evaluation and plan to conduct formal human evaluations for public models that perform well on the Vibe-Eval's automatic scores. We release the evaluation code and data, see https://github.com/reka-ai/reka-vibe-eval