annotations_creators	language	license	multilinguality	pretty_name	size_categories	source_datasets	tags	task_categories	dataset_info
expert-generated	en	apache-2.0	monolingual	ALL Bench Leaderboard 2026	n<1K	original	benchmark leaderboard llm vlm ai-evaluation gpt-5 claude gemini final-bench metacognition multimodal ai-agent image-generation video-generation music-generation	text-generation visual-question-answering text-to-image text-to-video text-to-audio	features name dtype llm list name dtype vlm dict name dtype agent list name dtype image list name dtype video list name dtype music list name dtype confidence dict

🏆 ALL Bench Leaderboard 2026

The only AI benchmark dataset covering LLM · VLM · Agent · Image · Video · Music in a single unified file.

Dataset Summary

ALL Bench Leaderboard aggregates and cross-verifies benchmark scores for 91 AI models across 6 modalities. Every numerical score is tagged with a confidence level (cross-verified, single-source, or self-reported) and its original source. The dataset is designed for researchers, developers, and decision-makers who need a trustworthy, unified view of the AI model landscape.

Category	Models	Benchmarks	Description
LLM	42	31 fields	MMLU-Pro, GPQA, AIME, HLE, ARC-AGI-2, Metacog, SWE-Pro, IFEval, LCB, etc.
VLM Flagship	11	10 fields	MMMU, MMMU-Pro, MathVista, AI2D, OCRBench, MMStar, HallusionBench, etc.
VLM Lightweight	5	34 fields	Detailed Qwen-series edge model comparison across 3 sub-categories
Agent	10	8 fields	OSWorld, τ²-bench, BrowseComp, Terminal-Bench 2.0, GDPval-AA, SWE-Pro
Image Gen	10	7 fields	Photo realism, text rendering, instruction following, style, aesthetics
Video Gen	10	7 fields	Quality, motion, consistency, text rendering, duration, resolution
Music Gen	8	6 fields	Quality, vocals, instrumental, lyrics, duration

Live Leaderboard

👉 https://huggingface.co/spaces/FINAL-Bench/all-bench-leaderboard

Interactive features: composite ranking, dark mode, advanced search (GPQA > 90 open, price < 1), Model Finder, Head-to-Head comparison, Trust Map heatmap, Bar Race animation, and downloadable Intelligence Report (PDF/DOCX).

Data Structure

all_bench_leaderboard_v2.1.json
├── metadata          # version, formula, links, model counts
├── llm[42]           # 42 LLMs × 31 fields
├── vlm
│   ├── flagship[11]  # 11 flagship VLMs × 10 benchmarks
│   └── lightweight[5]# 5 edge models × 34 benchmarks (3 sub-tables)
├── agent[10]         # 10 agent models × 8 benchmarks
├── image[10]         # 10 image gen models × S/A/B/C ratings
├── video[10]         # 10 video gen models × S/A/B/C ratings
├── music[8]          # 8 music gen models × S/A/B/C ratings
└── confidence{42}    # per-model, per-benchmark source & trust level

LLM Field Schema

Field	Type	Description
name	string	Model name
provider	string	Organization
type	string	open or closed
group	string	flagship, open, korean, etc.
released	string	Release date (YYYY.MM)
mmluPro	float | null	MMLU-Pro score (%)
gpqa	float | null	GPQA Diamond (%)
aime	float | null	AIME 2025 (%)
hle	float | null	Humanity's Last Exam (%)
arcAgi2	float | null	ARC-AGI-2 (%)
metacog	float | null	FINAL Bench Metacognitive score
swePro	float | null	SWE-bench Pro (%)
bfcl	float | null	Berkeley Function Calling (%)
ifeval	float | null	IFEval instruction following (%)
lcb	float | null	LiveCodeBench (%)
sweV	float | null	SWE-bench Verified (%) — deprecated
mmmlu	float | null	Multilingual MMLU (%)
termBench	float | null	Terminal-Bench 2.0 (%)
sciCode	float | null	SciCode (%)
priceIn / priceOut	float | null	USD per 1M tokens
elo	int | null	Arena Elo rating
license	string	Prop, Apache2, MIT, Open, etc.

Composite Score

Score = Avg(confirmed benchmarks) × √(N/10)

10 core benchmarks across the 5-Axis Intelligence Framework: Knowledge · Expert Reasoning · Abstract Reasoning · Metacognition · Execution.

Confidence System

Each benchmark score in the confidence object is tagged:

Level	Badge	Meaning
cross-verified	✓✓	Confirmed by 2+ independent sources
single-source	✓	One official or third-party source
self-reported	~	Provider's own claim, unverified

Example:

"Claude Opus 4.6": {
  "gpqa": { "level": "cross-verified", "source": "Anthropic + Vellum + DataCamp" },
  "arcAgi2": { "level": "cross-verified", "source": "Vellum + llm-stats + NxCode + DataCamp" },
  "metacog": { "level": "single-source", "source": "FINAL Bench dataset" }
}

Usage

import json
from huggingface_hub import hf_hub_download

path = hf_hub_download(
    repo_id="FINAL-Bench/ALL-Bench-Leaderboard",
    filename="all_bench_leaderboard_v2.1.json",
    repo_type="dataset"
)
data = json.load(open(path))

# Top 5 LLMs by GPQA
ranked = sorted(data["llm"], key=lambda x: x["gpqa"] or 0, reverse=True)
for m in ranked[:5]:
    print(f"{m['name']:25s} GPQA={m['gpqa']}")

# Check confidence for a score
print(data["confidence"]["Gemini 3.1 Pro"]["gpqa"])
# → {"level": "single-source", "source": "Google DeepMind model card"}

FINAL Bench — Metacognitive Benchmark

FINAL Bench measures AI self-correction ability. Error Recovery (ER) explains 94.8% of metacognitive performance variance. 9 frontier models evaluated.

• 🧬 FINAL-Bench/Metacognitive Dataset
• 🏆 FINAL-Bench/Leaderboard

Citation

@misc{allbench2026,
    title={ALL Bench Leaderboard 2026: Unified Multi-Modal AI Evaluation},
    author={ALL Bench Team},
    year={2026},
    url={https://huggingface.co/spaces/FINAL-Bench/all-bench-leaderboard}
}

---

#AIBenchmark #LLMLeaderboard #GPT5 #Claude #Gemini #ALLBench #FINALBench #Metacognition #VLM #AIAgent #MultiModal #HuggingFace #ARC-AGI #AIEvaluation #VIDRAFT.net