Start free
Tools · Category

ml frameworks, ranked by real usage_

As of 4 November 2026, PyTorch leads ml frameworks with 5,581 companies naming it in job postings, ahead of TensorFlow at 4,387.

As of 4 November 2026 · refreshed daily · counts move as websites, DNS and career sites are re-read.

Leader
PyTorch · 5,581
Runner-up
TensorFlow · 4,387
Most hiring, 90 days
PyTorch
Highest postings per company
Apache MXNet
Catalogue
T-1.0.1-prov
18,225companies across the category
31%of them on PyTorch
47,946postings naming a ml framework in 90 days
40tools with enough evidence to rank
02 — Ranking

Every ml framework we can measure

Companies naming the tool, and postings naming it in the last 90 days. Bars are relative to the leader.

#ToolCompaniescompaniesPostings, 90 dayspostingsper company
1
PyTorchPyTorch Foundation
5,581
17,5043.1
2
TensorFlowGoogle
4,387
12,0072.7
3
scikit-learnscikit-learn developers
1,639
3,8102.3
4
OpenCVOpenCV.org
812
1,5862.0
5
JAXGoogle
740
1,8782.5
6
Hugging Face TransformersHugging Face
616
1,2632.1
7
KerasGoogle
605
1,3202.2
8
XGBoostDMLC
574
1,0971.9
9
ONNXLinux Foundation
435
8351.9
10
LightGBMMicrosoft
267
5111.9
11
Core MLApple
214
4262.0
12
DeepSpeedMicrosoft
207
3281.6
13
Hugging Face PEFTHugging Face
205
5082.5
14
spaCyExplosion
193
3882.0
15
PyTorch FSDPMeta
163
2691.7
16
YOLOUltralytics
163
2811.7
17
NLTKNLTK Project
145
2781.9
18
TensorFlow LiteGoogle
127
5134.0
19
Megatron-LMNVIDIA
113
2842.5
20
NVIDIA NeMoNVIDIA
111
5795.2
21
SHAPSHAP
100
2152.1
22
Apache TVMApache Software Foundation
94
2562.7
23
XLAOpenXLA
79
2463.1
24
TRLHugging Face
61
961.6
25
Spark MLlibApache Software Foundation
60
1752.9
26
AWQ
52
1422.7
27
LIMELIME
50
1102.2
28
CatBoostYandex
47
942.0
29
CaffeBAIR
45
871.9
30
HALCONMVTec
42
1132.7
31
verl
40
1132.8
32
scikit-imagescikit-image
37
691.9
33
Amazon ComprehendAmazon
34
812.4
34
GensimRARE Technologies
34
541.6
35
FlashAttentionDao-AILab
28
923.3
36
Apache MXNetApache Software Foundation
27
1465.4
37
GymnasiumFarama Foundation
27
411.5
38
MediaPipeGoogle
25
241.0
39
IREEOpenXLA
23
532.3
40
MLXApple
23
743.2
03 — Head to head

Compare two tools

Same data, two columns: who chooses which by size, industry and country.

PyTorch vs TensorFlow5,581 · 4,387PyTorch vs scikit-learn5,581 · 1,639PyTorch vs OpenCV5,581 · 812PyTorch vs JAX5,581 · 740PyTorch vs Hugging Face Transformers5,581 · 616TensorFlow vs scikit-learn4,387 · 1,639TensorFlow vs OpenCV4,387 · 812TensorFlow vs JAX4,387 · 740TensorFlow vs Hugging Face Transformers4,387 · 616scikit-learn vs OpenCV1,639 · 812scikit-learn vs JAX1,639 · 740scikit-learn vs Hugging Face Transformers1,639 · 616OpenCV vs JAX812 · 740OpenCV vs Hugging Face Transformers812 · 616JAX vs Hugging Face Transformers740 · 616
Cite this

Quote the sentence with the date and the link. The numbers change daily; the link always shows the current ones.

PyTorch leads ml frameworks with 5,581 companies naming it in job postings, ahead of TensorFlow at 4,387. (Breaknorm, 4 November 2026, https://breaknorm.co/tools/category/ml-frameworks)
How we know

Read from the source, not self-reported

Every number on this page can be traced to a page we crawled or a posting we read. Full methodology →

  • CatalogueBreaknorm technology catalogue T-1.0.1-prov: 6,296 tools with vendor and category, editions folded into their parent.
  • CompaniesCompanies whose own postings name the tool as the team's system, any confidence. Candidate wish-lists excluded.
  • 90-day postingsOpen postings first seen or posted in the last 90 days that name the tool.
  • Not countedTools used by teams that do not post jobs, and companies without a career site we can read.
FAQ

ml frameworks: questions

How is this ranking made?

By the number of companies whose own job postings name the tool as the team's system, from the Breaknorm technology catalogue (version T-1.0.1-prov). No vendor paid for a position and no editor chose the order.

Why is this different from review-site rankings?

Review sites rank by reviews collected. This ranks by companies hiring people to work in the tool, which is closer to seats in use. It under-counts tools used by teams that rarely post jobs.

What about product editions?

Editions and sub-products () are counted under the parent so no company is counted twice.

Can I get the companies behind a number?

Yes: click a tool, see the first 20, and sign up for the full list with verified e-mails. Counting is free.

Companies and people

Get the companies your way

The same data as a search, over MCP and the API, as bulk files, or pushed to you as it changes.

SearchCompany searchFilter on what a company actually does→SearchPeople searchDecision makers, normalized and verified→MCP and APICompanies over MCP and APICount, search and export from your agent→DatasetCompany and People datasetThe whole market in your warehouse→DataCoverage by countryCheck your market before you start→SearchLocal businessesThe same database, by category and city→

Every company on every tool,
ranked by what they say

Connect the MCP server, ask which companies run a tool, and get 1,000 records free.

Start free with 1,000 records
  1. 01Sign up, no card
  2. 02Connect Claude, ChatGPT or an API key
  3. 03Ask for your market, 1,000 records free