If you search for companies hiring machine learning engineers, you mostly get lists of AI labs. That makes sense on the surface. OpenAI, Anthropic and Mistral are the companies everyone is talking about.
Count the job titles, though, and a different list comes out on top.
On October 6, 2026 we counted 11,704 open tech roles at 762 companies, as distinct roles (one role posted in ten cities counts once). We pulled every role whose title says "machine learning", "ML", "MLE" or "applied scientist". That gives 511 roles at 162 companies. About one open tech role in 23.
Who has the most ML engineer roles open
| Company | ML roles |
|---|---|
| 21 | |
| Nuro | 20 |
| 18 | |
| Roblox | 15 |
| Scale AI | 12 |
| Faire | 12 |
| Airbnb | 12 |
| Nebius | 11 |
| Lyft | 10 |
| Mistral AI | 10 |
| Databricks | 10 |
| Anthropic | 10 |
A social network, an autonomous driving company, a pinboard, a games platform, a wholesale marketplace and a home rental site, all ahead of the frontier labs. OpenAI has 9 roles with an ML title, which puts it just outside the top twelve.
This is not because the labs are not hiring. It is because they call the job something else.
The labs hire "research engineers"
Widen the net to titles with "research engineer" or "AI engineer" in them and the picture flips. That broader group is 1,283 roles at 321 companies, and now the labs lead the list: OpenAI 62, Anthropic 49, Mistral 35, Databricks 34.
"Research engineer" alone is 121 roles, and four companies hold 57 of them: Anthropic (29), OpenAI (11), Mistral (9) and Turing (8). If research engineering is what you are after, that is nearly half the market in four places.
So the practical advice is a vocabulary one. If you want classic ML engineering (ranking, recommendations, ads, fraud, perception), search "machine learning" and you will mostly land at product companies. If you want to work on models themselves, search "research engineer" and you will mostly land at labs. Same skills on paper, different job, different employers.
(The "AI engineer" titles are the loosest of all. Some of them are solutions or forward deployed roles that build on models rather than train them. We kept them out of the core count for that reason.)
What the stack looks like
Across the 511 core ML roles:
Python is in 332 (65%)
PyTorch in 195 (38%)
AWS in 112
TensorFlow in 99
Kubernetes in 95
SQL in 88
Spark in 78
PyTorch outnumbers TensorFlow about two to one. If you are deciding which one to get properly good at, the board has an opinion.
The research engineer roles look different: reinforcement learning shows up in 20 of them, JAX in 16, CUDA in 12. That is the infrastructure-of-training end of the field, and it is a much smaller and more specialised market.
Pay
325 of the core ML roles publish a yearly US dollar band. The median floor is $200,000 and the median ceiling is $290,000. For comparison, the median floor across all Python roles on the board is $168,650, so ML titles sit about $30,000 higher at the bottom of the band.
Research engineer roles pay more again: the 72 that publish a band average a $406,000 midpoint. But four labs hold nearly half of those roles, so we would not read that as the going rate for the skill.
Getting in, and where you can work from
Of the 511 core ML roles, 351 are senior (69%), 130 mid-level and 30 junior. That 5.9% junior share is actually a little above the 4.6% across all open tech roles. Not easy, but not the closed door it is sometimes made out to be.
Remote is another story. In the broader group of 1,283, 44% are hybrid, 30% are remote but restricted to a country or region, 24% are onsite, and 17 roles (1.3%) are labelled remote from anywhere. Of the restricted ones, about 40% are restricted to the US. ML engineering is overwhelmingly an office-or-near-it job right now.
How to search this market
Search titles for "machine learning" and "research engineer" separately. They are different markets.
Do not skip the product companies. Reddit, Pinterest, Roblox, Airbnb and Lyft have as many ML-titled roles open as most labs, or more.
Get PyTorch to a level you can talk about in an interview before you add a second framework.
Plan on hybrid. If you need fully remote, the ML market is the wrong place to start looking.
Every role on remoet.dev/jobs carries its stack and remote policy, and an agent can search it without an account. Back in June we looked at the same question from the company side, how many companies say "AI" against how many show a real ML stack, in who is actually hiring AI engineers.
Where this data is soft
Title matching is blunt. "ML" catches the odd title that is ML-adjacent rather than ML engineering, and it misses ML work hiding under titles like "Software Engineer, Ranking". The stack on each role is extracted by a model from the post, so "PyTorch or TensorFlow" counts for both. Salaries are only roles that publish a band. And it is one day, October 6, 2026.
If you hire ML engineers, or are one, and this does not match what you see, tell us in Discord. We would rather be corrected than quoted.
