DropTheHQ/benchgecko-ai-models

BenchGecko AI Model Benchmarks 2026

A comprehensive dataset of 413 AI models with benchmark scores across 40 evaluations, pricing data, and provider information. Sourced from BenchGecko, an independent AI model tracking platform.

Dataset Summary

Metric	Count
Models tracked	413
Benchmarks	40
Providers	57
Individual scores	1,577
Models with pricing	346
Open source models	235

Last updated: March 2026

Files

File	Description	Rows	Columns
models.csv	All models with metadata and benchmark scores	413	51
benchmarks.csv	All 40 benchmarks with categories and top performers	40	12
providers.csv	57 AI providers with model counts and avg pricing	57	7
scores.csv	Normalized model-benchmark score pairs (long format)	1,577	7

Column Descriptions

models.csv

Column	Type	Description
slug	string	Unique model identifier
name	string	Model display name
provider	string	Company/org that created the model
provider_slug	string	Provider identifier
avg_score	float	Average score across all benchmarks (0-100)
input_price_per_mtok	float	Input pricing in USD per million tokens
output_price_per_mtok	float	Output pricing in USD per million tokens
context_window	int	Maximum context window in tokens
is_open_source	bool	Whether model weights are publicly available
release_date	date	Model release date (YYYY-MM-DD)
model_type	string	Model modality (text, multimodal, etc.)
score_*	float	Individual benchmark scores (40 columns)

benchmarks.csv

Column	Type	Description
slug	string	Benchmark identifier
name	string	Benchmark display name
category	string	Category: knowledge, reasoning, math, coding, agentic, multimodal
max_score	float	Maximum possible score
unit	string	Score unit (typically %)
models_tested	int	Number of models evaluated
top_model_1/2/3	string	Top 3 performing models
top_score_1/2/3	float	Corresponding top scores

providers.csv

Column	Type	Description
slug	string	Provider identifier
name	string	Provider display name
model_count	int	Total models from this provider
models_with_pricing	int	Models with published pricing
avg_input_price_per_mtok	float	Average input price (USD/Mtok)
avg_output_price_per_mtok	float	Average output price (USD/Mtok)
open_source_count	int	Number of open source models

scores.csv

Long-format table for easy filtering and analysis:

Column	Type	Description
model_slug	string	Model identifier
model_name	string	Model name
provider	string	Provider name
benchmark_slug	string	Benchmark identifier
benchmark_name	string	Benchmark display name
score	float	Score value
benchmark_category	string	Benchmark category

Benchmark Categories

Category	Benchmarks	Description
Knowledge	17	Factual knowledge (MMLU, ARC, TriviaQA, etc.)
Math	5	Mathematical reasoning (GSM8K, MATH, FrontierMath)
Coding	7	Code generation/understanding (SWE-Bench, Aider, WeirdML)
Reasoning	3	Logical reasoning (BBH, SimpleBench, ARC-AGI-2)
Agentic	3	Agent capabilities (APEX-Agents, OSWorld, The Agent Company)
Multimodal	1	Video understanding (VideoMME)

Usage Examples

Load with pandas

import pandas as pd

# Load all models
models = pd.read_csv("hf://datasets/DropTheHQ/benchgecko-ai-models/models.csv")

# Top 10 models by average score
print(models.nlargest(10, "avg_score")[["name", "provider", "avg_score"]])

# Compare open source vs proprietary
print(models.groupby("is_open_source")["avg_score"].describe())

Filter by provider

# All OpenAI models sorted by avg score
openai = models[models["provider"] == "OpenAI"].sort_values("avg_score", ascending=False)
print(openai[["name", "avg_score", "input_price_per_mtok"]].head(10))

Benchmark analysis

scores = pd.read_csv("hf://datasets/DropTheHQ/benchgecko-ai-models/scores.csv")

# Best model per benchmark
best = scores.loc[scores.groupby("benchmark_slug")["score"].idxmax()]
print(best[["benchmark_name", "model_name", "score", "benchmark_category"]])

Price-performance analysis

# Models with pricing data
priced = models[models["input_price_per_mtok"].notna()].copy()
priced["cost_efficiency"] = priced["avg_score"] / priced["input_price_per_mtok"]
print(priced.nlargest(10, "cost_efficiency")[["name", "avg_score", "input_price_per_mtok", "cost_efficiency"]])

Load with Hugging Face datasets

from datasets import load_dataset

ds = load_dataset("DropTheHQ/benchgecko-ai-models")
print(ds)

Source

Data sourced from BenchGecko -- an independent platform that tracks AI models, agents, benchmarks, and pricing. BenchGecko aggregates benchmark results from official papers, evaluation suites, and reproducible third-party tests.

Explore the full interactive leaderboard at benchgecko.ai.

License

This dataset is released under CC BY 4.0. You are free to share and adapt this data for any purpose, including commercial, as long as you give appropriate credit.

Citation

@dataset{benchgecko_ai_models_2026,
  title={BenchGecko AI Model Benchmarks 2026},
  author={BenchGecko},
  year={2026},
  url={https://benchgecko.ai},
  publisher={Hugging Face},
  license={CC BY 4.0},
  note={413 AI models, 40 benchmarks, 57 providers, 1577 scores}
}