Dataset Preview
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
The dataset generation failed because of a cast error
Error code: DatasetGenerationCastError
Exception: DatasetGenerationCastError
Message: An error occurred while generating the dataset
All the data files must have the same columns, but at some point there are 6 new columns ({'github_stars', 'name', 'github_url', 'author', 'last_updated', 'language'}) and 4 missing columns ({'models_tested', 'max_score', 'methodology', 'benchmark'}).
This happened while the csv dataset builder was generating data using
hf://datasets/BenchGeckoAI/ai-model-benchmarks-2026/data/mcp_servers.csv (at revision 59b53f222028a66ddf45cb132b15505923bb69ed), [/tmp/hf-datasets-cache/medium/datasets/25809192754235-config-parquet-and-info-BenchGeckoAI-ai-model-ben-fab6395f/hub/datasets--BenchGeckoAI--ai-model-benchmarks-2026/snapshots/59b53f222028a66ddf45cb132b15505923bb69ed/data/benchmarks.csv (origin=hf://datasets/BenchGeckoAI/ai-model-benchmarks-2026@59b53f222028a66ddf45cb132b15505923bb69ed/data/benchmarks.csv), /tmp/hf-datasets-cache/medium/datasets/25809192754235-config-parquet-and-info-BenchGeckoAI-ai-model-ben-fab6395f/hub/datasets--BenchGeckoAI--ai-model-benchmarks-2026/snapshots/59b53f222028a66ddf45cb132b15505923bb69ed/data/mcp_servers.csv (origin=hf://datasets/BenchGeckoAI/ai-model-benchmarks-2026@59b53f222028a66ddf45cb132b15505923bb69ed/data/mcp_servers.csv), /tmp/hf-datasets-cache/medium/datasets/25809192754235-config-parquet-and-info-BenchGeckoAI-ai-model-ben-fab6395f/hub/datasets--BenchGeckoAI--ai-model-benchmarks-2026/snapshots/59b53f222028a66ddf45cb132b15505923bb69ed/data/models.csv (origin=hf://datasets/BenchGeckoAI/ai-model-benchmarks-2026@59b53f222028a66ddf45cb132b15505923bb69ed/data/models.csv), /tmp/hf-datasets-cache/medium/datasets/25809192754235-config-parquet-and-info-BenchGeckoAI-ai-model-ben-fab6395f/hub/datasets--BenchGeckoAI--ai-model-benchmarks-2026/snapshots/59b53f222028a66ddf45cb132b15505923bb69ed/data/providers.csv (origin=hf://datasets/BenchGeckoAI/ai-model-benchmarks-2026@59b53f222028a66ddf45cb132b15505923bb69ed/data/providers.csv)]
Please either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.12/site-packages/datasets/builder.py", line 1893, in _prepare_split_single
writer.write_table(table)
File "/usr/local/lib/python3.12/site-packages/datasets/arrow_writer.py", line 765, in write_table
self._write_table(pa_table, writer_batch_size=writer_batch_size)
File "/usr/local/lib/python3.12/site-packages/datasets/arrow_writer.py", line 773, in _write_table
pa_table = table_cast(pa_table, self._schema)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2281, in table_cast
return cast_table_to_schema(table, schema)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2227, in cast_table_to_schema
raise CastError(
datasets.table.CastError: Couldn't cast
name: string
slug: string
category: string
description: string
github_stars: int64
github_url: string
author: string
language: string
last_updated: string
-- schema metadata --
pandas: '{"index_columns": [{"kind": "range", "name": null, "start": 0, "' + 1307
to
{'benchmark': Value('string'), 'slug': Value('string'), 'category': Value('string'), 'description': Value('string'), 'max_score': Value('int64'), 'models_tested': Value('int64'), 'methodology': Value('string')}
because column names don't match
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1347, in compute_config_parquet_and_info_response
parquet_operations = convert_to_parquet(builder)
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 980, in convert_to_parquet
builder.download_and_prepare(
File "/usr/local/lib/python3.12/site-packages/datasets/builder.py", line 884, in download_and_prepare
self._download_and_prepare(
File "/usr/local/lib/python3.12/site-packages/datasets/builder.py", line 947, in _download_and_prepare
self._prepare_split(split_generator, **prepare_split_kwargs)
File "/usr/local/lib/python3.12/site-packages/datasets/builder.py", line 1739, in _prepare_split
for job_id, done, content in self._prepare_split_single(
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/builder.py", line 1895, in _prepare_split_single
raise DatasetGenerationCastError.from_cast_error(
datasets.exceptions.DatasetGenerationCastError: An error occurred while generating the dataset
All the data files must have the same columns, but at some point there are 6 new columns ({'github_stars', 'name', 'github_url', 'author', 'last_updated', 'language'}) and 4 missing columns ({'models_tested', 'max_score', 'methodology', 'benchmark'}).
This happened while the csv dataset builder was generating data using
hf://datasets/BenchGeckoAI/ai-model-benchmarks-2026/data/mcp_servers.csv (at revision 59b53f222028a66ddf45cb132b15505923bb69ed), [/tmp/hf-datasets-cache/medium/datasets/25809192754235-config-parquet-and-info-BenchGeckoAI-ai-model-ben-fab6395f/hub/datasets--BenchGeckoAI--ai-model-benchmarks-2026/snapshots/59b53f222028a66ddf45cb132b15505923bb69ed/data/benchmarks.csv (origin=hf://datasets/BenchGeckoAI/ai-model-benchmarks-2026@59b53f222028a66ddf45cb132b15505923bb69ed/data/benchmarks.csv), /tmp/hf-datasets-cache/medium/datasets/25809192754235-config-parquet-and-info-BenchGeckoAI-ai-model-ben-fab6395f/hub/datasets--BenchGeckoAI--ai-model-benchmarks-2026/snapshots/59b53f222028a66ddf45cb132b15505923bb69ed/data/mcp_servers.csv (origin=hf://datasets/BenchGeckoAI/ai-model-benchmarks-2026@59b53f222028a66ddf45cb132b15505923bb69ed/data/mcp_servers.csv), /tmp/hf-datasets-cache/medium/datasets/25809192754235-config-parquet-and-info-BenchGeckoAI-ai-model-ben-fab6395f/hub/datasets--BenchGeckoAI--ai-model-benchmarks-2026/snapshots/59b53f222028a66ddf45cb132b15505923bb69ed/data/models.csv (origin=hf://datasets/BenchGeckoAI/ai-model-benchmarks-2026@59b53f222028a66ddf45cb132b15505923bb69ed/data/models.csv), /tmp/hf-datasets-cache/medium/datasets/25809192754235-config-parquet-and-info-BenchGeckoAI-ai-model-ben-fab6395f/hub/datasets--BenchGeckoAI--ai-model-benchmarks-2026/snapshots/59b53f222028a66ddf45cb132b15505923bb69ed/data/providers.csv (origin=hf://datasets/BenchGeckoAI/ai-model-benchmarks-2026@59b53f222028a66ddf45cb132b15505923bb69ed/data/providers.csv)]
Please either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
benchmark string | slug string | category string | description string | max_score int64 | models_tested int64 | methodology string |
|---|---|---|---|---|---|---|
MMLU | mmlu | Knowledge | Massive Multitask Language Understanding - 57 subjects from STEM to humanities | 100 | 380 | multiple_choice |
MMLU-Pro | mmlu-pro | Knowledge | Harder MMLU with 10 answer choices and more reasoning-heavy questions | 100 | 245 | multiple_choice |
HumanEval | humaneval | Coding | Python function completion from docstrings - 164 problems | 100 | 340 | pass_at_1 |
MBPP | mbpp | Coding | Mostly Basic Programming Problems - 974 crowd-sourced Python tasks | 100 | 310 | pass_at_1 |
SWE-bench Verified | swe-bench-verified | Coding | Real GitHub issue resolution on popular Python repos | 100 | 180 | execution |
GPQA Diamond | gpqa-diamond | Science | Graduate-level questions in physics chemistry and biology | 100 | 290 | multiple_choice |
MATH | math | Mathematics | Competition mathematics across 7 difficulty levels | 100 | 350 | exact_match |
GSM8K | gsm8k | Mathematics | Grade school math word problems - 8.5K examples | 100 | 370 | exact_match |
ARC-Challenge | arc-challenge | Reasoning | AI2 Reasoning Challenge - grade-school science questions | 100 | 360 | multiple_choice |
HellaSwag | hellaswag | Reasoning | Sentence completion requiring commonsense reasoning | 100 | 355 | multiple_choice |
WinoGrande | winogrande | Reasoning | Pronoun resolution requiring world knowledge | 100 | 340 | accuracy |
TruthfulQA | truthfulqa | Safety | Measures tendency to generate false but plausible answers | 100 | 320 | mc_accuracy |
BBH | bbh | Reasoning | BIG-Bench Hard - 23 challenging tasks from BIG-Bench | 100 | 300 | exact_match |
DROP | drop | Reading | Discrete Reasoning Over Paragraphs - reading comprehension with math | 100 | 280 | f1_score |
MGSM | mgsm | Multilingual | Multilingual Grade School Math in 10 languages | 100 | 220 | exact_match |
IFEval | ifeval | Instruction | Instruction Following Evaluation - verifiable instruction constraints | 100 | 260 | strict_accuracy |
MuSR | musr | Reasoning | Multi-Step Soft Reasoning - complex multi-hop problems | 100 | 200 | accuracy |
MMMU | mmmu | Multimodal | Massive Multi-discipline Multimodal Understanding | 100 | 120 | accuracy |
LiveCodeBench | livecodebench | Coding | Contamination-free coding benchmark from recent competitions | 100 | 190 | pass_at_1 |
Aider Polyglot | aider-polyglot | Coding | Multi-language code editing benchmark using Aider framework | 100 | 160 | edit_accuracy |
Arena ELO | arena-elo | Human Preference | Chatbot Arena crowdsourced human preference ratings | 2,000 | 280 | elo_rating |
MT-Bench | mt-bench | Conversation | Multi-turn conversation quality scored by GPT-4 | 10 | 310 | gpt4_judge |
AlpacaEval 2.0 | alpacaeval-2 | Instruction | Length-controlled win rate against GPT-4 Turbo baseline | 100 | 250 | lc_win_rate |
SimpleQA | simpleqa | Factuality | Short-form factual question answering with verifiable answers | 100 | 230 | exact_match |
BFCL | bfcl | Tool Use | Berkeley Function Calling Leaderboard - API and tool use | 100 | 180 | ast_accuracy |
Tau-bench | tau-bench | Agentic | Real-world agent task completion across airline and retail domains | 100 | 140 | task_success |
WebArena | webarena | Agentic | Web browsing agent tasks on realistic websites | 100 | 90 | task_completion |
AIME 2024 | aime-2024 | Mathematics | American Invitational Mathematics Examination problems | 100 | 200 | exact_match |
AMC 2023 | amc-2023 | Mathematics | American Mathematics Competition problems | 100 | 220 | exact_match |
Codeforces | codeforces | Coding | Competitive programming problems rated by difficulty | 3,000 | 150 | elo_rating |
RULER | ruler | Long Context | Synthetic long-context retrieval and reasoning tasks | 100 | 110 | accuracy |
NIAH | niah | Long Context | Needle in a Haystack - information retrieval in long documents | 100 | 130 | recall |
LongBench | longbench | Long Context | Real-world long document understanding tasks | 100 | 120 | f1_score |
Chatbot Arena Hard | arena-hard | Human Preference | Challenging subset of Arena prompts with high separability | 100 | 210 | win_rate |
NATURAL | natural | Tool Use | Natural language to API call translation | 100 | 160 | accuracy |
ToolBench | toolbench | Tool Use | Large-scale tool use benchmark with 16K real APIs | 100 | 130 | pass_rate |
MedQA | medqa | Domain | US Medical Licensing Exam style questions | 100 | 240 | accuracy |
LegalBench | legalbench | Domain | Legal reasoning across 162 tasks | 100 | 180 | accuracy |
FinanceBench | financebench | Domain | Financial analysis and reasoning from SEC filings | 100 | 150 | accuracy |
CLRS | clrs | Reasoning | Classical algorithm reasoning and step tracing | 100 | 100 | accuracy |
null | filesystem | developer-tools | Read write and manage local filesystem operations via MCP | null | null | null |
null | github | developer-tools | Interact with GitHub repositories issues and pull requests | null | null | null |
null | postgres | databases | Query and manage PostgreSQL databases with schema introspection | null | null | null |
null | brave-search | search | Web search via Brave Search API with summarization | null | null | null |
null | puppeteer | web-automation | Browser automation for web scraping and testing via Puppeteer | null | null | null |
null | slack | communication | Read and send Slack messages manage channels and users | null | null | null |
null | google-maps | location | Geocoding directions and place search via Google Maps API | null | null | null |
null | sentry | monitoring | Access Sentry error tracking events and issue data | null | null | null |
null | memory | knowledge | Persistent knowledge graph for long-term memory storage | null | null | null |
null | sqlite | databases | Query and manage SQLite databases with full SQL support | null | null | null |
null | fetch | web-automation | HTTP requests to any URL with response parsing and caching | null | null | null |
null | sequential-thinking | reasoning | Step-by-step reasoning and problem decomposition tool | null | null | null |
null | notion | productivity | Read and write Notion pages databases and blocks | null | null | null |
null | linear | project-management | Manage Linear issues projects and cycles via MCP | null | null | null |
null | supabase | databases | Query and manage Supabase projects databases and auth | null | null | null |
null | vercel | deployment | Manage Vercel deployments projects and environment variables | null | null | null |
null | stripe | payments | Access Stripe payment data customers and subscriptions | null | null | null |
null | jira | project-management | Manage Jira issues sprints and project boards | null | null | null |
null | docker | devops | Manage Docker containers images and compose stacks | null | null | null |
null | aws | cloud | Interact with AWS services including S3 Lambda and DynamoDB | null | null | null |
null | gpt-4o | null | null | null | null | null |
null | gpt-4o-mini | null | null | null | null | null |
null | gpt-4-5 | null | null | null | null | null |
null | o3 | null | null | null | null | null |
null | o4-mini | null | null | null | null | null |
null | claude-sonnet-4 | null | null | null | null | null |
null | claude-opus-4 | null | null | null | null | null |
null | claude-3-5-haiku | null | null | null | null | null |
null | claude-opus-4-6 | null | null | null | null | null |
null | gemini-2-0-flash | null | null | null | null | null |
null | gemini-2-5-pro | null | null | null | null | null |
null | gemini-2-5-flash | null | null | null | null | null |
null | llama-4-maverick | null | null | null | null | null |
null | llama-4-scout | null | null | null | null | null |
null | deepseek-v3 | null | null | null | null | null |
null | deepseek-r1 | null | null | null | null | null |
null | grok-3 | null | null | null | null | null |
null | grok-3-mini | null | null | null | null | null |
null | mistral-large-2 | null | null | null | null | null |
null | command-r-plus | null | null | null | null | null |
null | openai | null | null | null | null | null |
null | anthropic | null | null | null | null | null |
null | google | null | null | null | null | null |
null | meta | null | null | null | null | null |
null | deepseek | null | null | null | null | null |
null | xai | null | null | null | null | null |
null | mistral | null | null | null | null | null |
null | cohere | null | null | null | null | null |
null | amazon | null | null | null | null | null |
null | ai21-labs | null | null | null | null | null |
null | alibaba | null | null | null | null | null |
null | zhipu-ai | null | null | null | null | null |
null | 01-ai | null | null | null | null | null |
null | nvidia | null | null | null | null | null |
null | reka | null | null | null | null | null |
null | writer | null | null | null | null | null |
null | inflection | null | null | null | null | null |
null | databricks | null | null | null | null | null |
null | together-ai | null | null | null | null | null |
null | perplexity | null | null | null | null | null |
AI Model Benchmarks & Pricing Dataset 2026
A comprehensive survey of large language model performance and economics, maintained by BenchGecko.
What's Inside
| File | Records | Description |
|---|---|---|
data/models.csv |
20 | Top AI models with benchmark scores and API pricing |
data/providers.csv |
20 | AI model providers with metadata |
data/benchmarks.csv |
40 | Benchmark suites with methodology |
data/mcp_servers.csv |
20 | Model Context Protocol servers |
This is a sample from the full dataset. The complete dataset covers thousands of models, hundreds of providers, and over a hundred benchmarks, updated every two hours at benchgecko.ai.
Fields (models.csv)
| Column | Type | Description |
|---|---|---|
name |
String | Model display name |
provider |
String | Organization that created the model |
input_price |
Float | USD per 1M input tokens |
output_price |
Float | USD per 1M output tokens |
context_window |
Integer | Maximum context length in tokens |
average_score |
Float | Weighted average across all benchmarks (0-100) |
mmlu_score |
Float | MMLU benchmark score |
humaneval_score |
Float | HumanEval coding score |
gpqa_score |
Float | GPQA Diamond science score |
math_score |
Float | MATH competition score |
open_source |
Boolean | Whether weights are publicly available |
release_date |
Date | Public release date |
Quick Start
from datasets import load_dataset
dataset = load_dataset("BenchGeckoAI/ai-model-benchmarks-2026")
models = dataset["train"]
# Find the best open-source model
open_models = [m for m in models if m["open_source"]]
best = max(open_models, key=lambda m: m["average_score"])
print(f"Best open model: {best['name']} ({best['average_score']})")
Use Cases
- Model Selection: Compare benchmark scores across evaluation types before deploying
- Cost Analysis: Find the best price-to-performance ratio across providers
- Market Research: Track the AI model landscape and provider ecosystem
- Academic Research: Study capability trajectories and scaling laws
Full Dataset
This sample covers 20 models. The full live dataset is available through:
- Web: BenchGecko Model Rankings
- API: BenchGecko API Documentation
- Pricing: Cross-Provider Pricing Comparison
- Compare: Side-by-Side Model Comparison
- Economy: AI Economy Dashboard
- Compute: AI Compute Supply Chain
- Mindshare: Developer Mindshare Arena
Methodology
Benchmark scores sourced from original model technical reports and cross-verified using open-source evaluation frameworks (EleutherAI lm-evaluation-harness, BigCode HumanEval+). Pricing collected from official API documentation, updated within 48 hours of changes.
Citation
@dataset{benchgecko2026,
author = {BenchGecko},
title = {AI Model Benchmarks and Pricing Dataset 2026},
year = {2026},
publisher = {Hugging Face},
url = {https://huggingface.co/datasets/BenchGeckoAI/ai-model-benchmarks-2026}
}
License
CC BY 4.0. Attribution: BenchGecko (https://benchgecko.ai)
- Downloads last month
- 61