cff-version: 1.2.0
title: "ALL Bench Leaderboard 2026: Unified Multi-Modal AI Evaluation"
message: "If you use this dataset, please cite it as below."
type: dataset
authors:
  - name: "ALL Bench Team"
url: "https://huggingface.co/spaces/FINAL-Bench/all-bench-leaderboard"
repository-code: "https://github.com/final-bench/ALL-Bench-Leaderboard"
license: MIT
version: "2.1"
date-released: "2026-03-08"
keywords:
  - ai-benchmark
  - llm-leaderboard
  - vlm
  - multimodal-ai
  - metacognition
  - final-bench
  - gpt-5
  - claude
  - gemini
abstract: >-
  ALL Bench Leaderboard is the only AI benchmark covering LLM, VLM, Agent,
  Image, Video, and Music generation in a single unified view. It cross-verifies
  91 AI models across 6 modalities with a 3-tier confidence system. Features
  composite 5-axis scoring (Knowledge, Expert Reasoning, Abstract Reasoning,
  Metacognition, Execution), interactive comparison tools, and downloadable
  intelligence reports. Includes FINAL Bench metacognitive evaluation where
  Error Recovery explains 94.8% of self-correction performance variance.