Tiered Team
OpenAI's median base is $250k. Anthropic's is $188k. The AI pay ladder, from the filings
Everyone talks about AI lab pay like it's a single number. It isn't. When you line up the base salaries these companies actually put on paper with the US government, you get a ladder - and a couple of surprises.
Quick version: OpenAI sits at the top with a median base of $250,000. Meta beats Google. And Anthropic, the lab everyone assumes pays the most, comes in at $188,000 - level with NVIDIA.
Before anyone shares that last line out of context, read on. There's a good reason for it.
The ladder
Every year, employers file salary paperwork with the US Department of Labor when they hire certain roles. It lists the job, the location and the base pay they commit to. We pulled every certified full-time filing decided between Oct 1, 2025 and Sep 30, 2026 - 494,766 of them - and grouped them by employer.
| Employer | Filings | Median base | Middle half |
|---|---|---|---|
| Mistral AI | 55 | $280,000 | $205,000 - $313,500 |
| OpenAI | 308 | $250,000 | $245,000 - $321,300 |
| xAI | 58 | $245,000 | $202,500 - $300,000 |
| Meta | 6,160 | $214,000 | $186,100 - $234,200 |
| 8,977 | $197,000 | $167,700 - $222,000 | |
| NVIDIA | 2,976 | $188,300 | $168,000 - $216,100 |
| Anthropic | 264 | $188,000 | $161,500 - $213,500 |
| Microsoft | 6,134 | $177,400 | $159,800 - $198,100 |
"Middle half" is the 25th to 75th percentile - where most offers land.
A few things jump out.
OpenAI's floor is high. A quarter of its filings are at $245,000 or less. That's barely below the median. OpenAI tends to file one number per role, and that number is big. Its software engineers come in at a median of $252,500 (94 filings) and its hardware engineers at $380,000 (14 filings - small, but eye-catching).
Mistral and xAI are paying to compete. Mistral's US arm is small - 55 filings, mostly research scientists in Palo Alto - but those research roles have a median of $280,000. xAI's 58 filings have a median of $245,000. When you're the challenger, you pay up.
Meta is the Big Tech pay leader. Across 6,160 filings Meta's median is $214,000, ahead of Google at $197,000 and Microsoft at $177,400. Meta's software engineers alone sit at $218,400.
Google DeepMind is hiding inside Google. DeepMind's US hires file under Google LLC, so we can't split them out. Google's research scientists come in at $210,000 across 337 filings.
So why is Anthropic "only" $188k?
This is the bit worth understanding, because it's true of a lot of offers you'll see.
Many employers don't file one number. They file a range - say $190,000 to $400,000. Our data uses the bottom of that range, because it's the only figure the employer is actually committing to.
Anthropic files ranges almost every time. An independent analysis by Zelcast found a range on 94% of Anthropic's engineering and research filings, and that many of those ranges start at the government-set minimum for that job and city (the "prevailing wage"). Using the midpoint of the range instead, Zelcast puts Anthropic's engineers at a median of $281,756.
So Anthropic isn't paying less than NVIDIA. It's showing you a wider window. The floor is $188k. Where you land in the range is a negotiation.
Its software engineers have a median floor of $213,500 (129 filings), which is right in line with Meta. Research roles are similar at $209,100, though on a smaller sample (29 filings).
What this means if you're weighing an offer
Three things worth keeping in mind.
- Base is the boring bit. These numbers don't include stock or bonus. At the labs, equity is usually the larger part of the package. A $250k base with fast-vesting stock and a $250k base with illiquid options are very different offers.
- Ask where you sit in the range. If a company files ranges, the bottom is a floor, not a target. Ask the recruiter directly where your offer sits and why.
- Volume matters. OpenAI, Anthropic and xAI filed 630 between them. Google filed 8,977 on its own. The labs are hiring a lot fewer people than the headlines suggest, so the bar is high.
If you want to compare yourself properly, start with the role, not the company. The median base for software developers across the whole dataset is $146,800, and for research scientists it's $190,900. In San Francisco, software developers come in at $202,000.
Jump or stay?
The ladder tells you where the money is. It doesn't tell you whether moving is right for you - that depends on your level, your growth and how you're valued where you are now.
That's what we built Tiered for. Get your Jump or Stay score - it takes a few minutes and shows you how your pay and progression stack up.
Tell us what you'd like us to dig into next.
Methodology: Based on 494,766 salary filings employers submitted to the US Department of Labor for fiscal year 2026 (certified, full-time; decided Oct 1, 2025 - Sep 30, 2026). Figures are the base pay offered - the lower bound where a range was filed - not bonus or equity. These are H-1B, H-1B1 and E-3 Labor Condition Applications, so they cover sponsored roles, not every hire. Source: US Department of Labor, Office of Foreign Labor Certification. Browse every company at tiered.app/salaries.