Independent ML Benchmark Tracking

State of the Art,
Verified.

17 research areas. 286+ benchmarks. Every result linked to its paper and code. No marketing claims — just data.

Explore BenchmarksAI Building Blocks
286+
Benchmarks
17
Research areas
143
Models tracked

Research Areas

Browse benchmarks by domain

Computer Vision

10 tasks

Detection, segmentation, classification, OCR

NLP

9 tasks

Language models, QA, translation, NER

Reasoning

MATH, GSM8K

Mathematical, logical, commonsense

Code

6 tasks

Generation, SWE-bench, debugging

Speech

5 tasks

ASR, TTS, speaker verification

Medical

4 tasks

Imaging, diagnosis, clinical NLP

Multimodal

5 tasks

Vision-language, VQA, text-to-image

Agentic AI

5 tasks

Autonomous agents, HCAST, time horizon

View all 17 areas →

Featured

164+ models

OCR Benchmarks

164+ models compared across OmniDocBench, OCRBench v2, and olmOCR. Open source vs vendor APIs.

74% resolved

SWE-bench SOTA

Which AI agents actually solve real GitHub issues? Latest verified scores.

56+ datasets

Text Embeddings (MTEB)

Deep dive into the MTEB leaderboard. KaLM, Qwen3, Gemini, OpenAI compared across 8 task categories.

40+ blocks

AI Building Blocks

Input-to-output transformations. Find the right model for your task.

7 categories

AI Arena Rankings

Live human preference rankings across text, code, vision, search, and image generation.

3 deep-dives

Academy

Learn ML by reproducing real benchmarks. Theory → reproduce → improve → leaderboard.

Why this exists

When Meta shut down Papers with Code in July 2025, the ML community lost its reference for what state-of-the-art looks like. 9,327 benchmarks, 79,817 papers, gone overnight.

CodeSOTA rebuilds that infrastructure — independently. We verify results ourselves where possible, link every claim to its source paper and code, and publish all data as open JSON. No corporate owner that might pull the plug.

"Outstanding work. Just yesterday I was searching for good OCR comparisons and found only marketing BS. Good job!"

AI Consultant — Voice-AI at scale

"Super clean, slop-free UI, but most importantly the copy: very precise positioning and project overview."

Senior Architect

Open data

All benchmark data available as JSON. No API key, no rate limits. Build dashboards, cite in papers, integrate into your tools.

Download JSONMethodology

Cite CodeSOTA

@misc{wikiel2025codesota,
  author = {Wikieł, Kacper},
  title = {CodeSOTA: Independent ML Benchmark Tracking},
  year = {2025},
  url = {https://codesota.com},
  note = {Accessed: 2025}
}

FAQ

What is CodeSOTA?

An independent ML benchmark tracking platform. Verified state-of-the-art results across 17 research areas — vision, NLP, reasoning, code, speech, medical AI, robotics, and more.

Is this a Papers with Code replacement?

We build on the PWC legacy after Meta shut it down in July 2025. 286+ benchmarks with links to implementations. Full story.

Are these benchmarks verified?

Yes. We run benchmarks independently where possible. All data includes source URLs and access dates. See methodology.

Can I use this data?

Yes. All data is open JSON at /data/benchmarks.json. Build dashboards, cite in papers, or integrate into your tools.

Stay current

New benchmarks and model comparisons, delivered occasionally.