# ALL Bench Leaderboard 2026

> The only AI benchmark leaderboard covering LLM, VLM, Agent, Image, Video, and Music generation in one place.

## Overview

ALL Bench Leaderboard is a unified multi-modal AI evaluation platform. It cross-verifies 91 AI models across 6 modalities (LLM, VLM, Agent, Image Generation, Video Generation, Music Generation) with a 3-tier confidence system. Every score is traceable to its original source.

## Key Facts

- Version: 2.1 (March 2026)
- Total Models: 42 LLMs + 11 VLMs + 10 Agents + 10 Image + 10 Video + 8 Music = 91
- Core Benchmarks: MMLU-Pro, GPQA, AIME, HLE, ARC-AGI-2, FINAL Bench (Metacognition), SWE-Pro, BFCL, IFEval, LiveCodeBench
- Scoring: Composite = Avg(verified benchmarks) × √(N/10)
- 5-Axis Framework: Knowledge, Expert Reasoning, Abstract Reasoning, Metacognition, Execution
- Confidence Levels: Cross-verified (2+ sources), Single-source, Self-reported
- License: MIT

## Links

- Live Leaderboard: https://huggingface.co/spaces/FINAL-Bench/all-bench-leaderboard
- Dataset: https://huggingface.co/datasets/FINAL-Bench/ALL-Bench-Leaderboard
- GitHub: https://github.com/final-bench/ALL-Bench-Leaderboard
- FINAL Bench Dataset: https://huggingface.co/datasets/FINAL-Bench/Metacognitive
- FINAL Bench Leaderboard: https://huggingface.co/spaces/FINAL-Bench/Leaderboard

## Top Models (March 2026)

### LLM Top 5 by Composite Score
1. Gemini 3.1 Pro (Google) — GPQA 94.3, ARC-AGI-2 77.1%
2. GPT-5.2 (OpenAI) — GPQA 93.2, AIME 100
3. Claude Opus 4.6 (Anthropic) — SWE-V 80.8, MMMU-Pro 85.1
4. Grok 4 Heavy (xAI) — GPQA 92.0, ARC-AGI-2 67.5%
5. Kimi K2.5 (Moonshot) — HLE 44.9, Metacog 68.71

### VLM Top 3 by MMMU
1. Gemini 3 Flash — MMMU 87.6%
2. Gemini 3 Pro — MMMU 87.5%
3. GPT-5.2 — MMMU 86.7%

### FINAL Bench Metacognitive Top 3
1. Kimi K2.5 — 68.71
2. GPT-5.2 — 62.76
3. GLM-5 — 62.50

## FINAL Bench

FINAL Bench (Frontier Intelligence Nexus for AGI-Level Verification) measures AI self-correction ability. Error Recovery (ER) explains 94.8% of metacognitive performance variance. 9 frontier models evaluated. Featured in Seoul Shinmun, Asia Economy, IT Chosun (2026.02.27). HuggingFace Datasets global ranking: Top 5.

## API

Free Gradio API with 8 endpoints. No authentication required.
Endpoints: /get_llm_data, /get_vlm_data, /get_agent_data, /get_image_data, /get_video_data, /get_music_data, /get_all_data, /search_models

## Data Format

Single unified JSON file: all_bench_leaderboard_v2.1.json (75KB)
Categories: llm[42], vlm.flagship[11], vlm.lightweight[5], agent[10], image[10], video[10], music[8], confidence{42 models}