ML engineer salary in 2026
ML engineer salary in 2026
The 2026 median machine learning engineer salary in the US is roughly $148,000 base and $212,000 total compensation for a generalist production MLE (Built In, Glassdoor). At FAANG and frontier labs, total comp lands much higher: $430k median at Meta, $290k at Google, ~$795k median at OpenAI, $300k to $490k at Anthropic (Levels.fyi). The number you should actually budget for depends on which MLE you are hiring, because "ML engineer" has fractured into at least three distinct jobs that pay very different amounts.
This guide breaks the market down honestly: the three MLE archetypes, real 2026 numbers by region, what the comp tables hide, and how the math changes when you book by the week instead of running a 6-month hiring loop.
The three ML engineer archetypes (and why one number doesn't work)
Almost every salary guide treats "ML engineer" as one role. In 2026 it is three:
- Frontier-lab researcher / research engineer. Pre-training, post-training, RLHF, capabilities work at OpenAI, Anthropic, Google DeepMind, xAI, Mistral. Comp is uncapped because each engineer's marginal contribution to a frontier model is measured in revenue points.
- FAANG production MLE. Ranking, ads, recommendations, fraud, search at Meta, Google, Amazon, Apple, LinkedIn, Snap. Big-tech band, big-tech equity, predictable ladders.
- Startup applied MLE. Ships features that wrap an LLM API, fine-tunes a Llama variant, runs a vector DB, builds eval pipelines. Lives in product code more than in research notebooks.
The compensation gap between archetype 1 and archetype 3 is roughly 5x to 10x for the same number of years of experience. Treating them as one number is how founders end up paying $300k for work a $1,500-per-week senior could ship in a month.
US 2026 numbers by archetype and level
Real ranges, sourced from Levels.fyi, Glassdoor, Built In, and BLS where possible.
Startup applied MLE (US, total comp)
| Level | Base | Equity (annualized) | Total comp |
|---|---|---|---|
| Junior (0-2 yrs) | $115k - $135k | $10k - $25k | $125k - $160k |
| Mid (3-5 yrs) | $140k - $175k | $25k - $50k | $165k - $225k |
| Senior (5-8 yrs) | $175k - $215k | $50k - $100k | $225k - $315k |
| Staff (8+ yrs) | $215k - $260k | $100k - $200k | $315k - $460k |
FAANG production MLE (median total comp, Levels.fyi 2026)
| Company | E3 / Junior | E4 / Mid | E5 / Senior | E6 / Staff |
|---|---|---|---|---|
| $199k | $290k | $410k | $580k | |
| Meta | $187k | $310k | $430k | $660k |
| Amazon | $176k | $230k | $310k | $399k |
| Apple | $190k | $260k | $355k | $470k |
| $246k | $360k | $490k | $690k | |
| Snap | n/a | $310k | $450k | $620k |
Frontier-lab MLE / research engineer (US, total comp)
| Lab | Junior research / SWE | Senior research engineer | Staff / member of technical staff |
|---|---|---|---|
| OpenAI | $249k - $400k | $530k - $900k | $1.0M - $3.2M+ |
| Anthropic | $300k - $400k | $400k - $550k | $550k - $900k |
| Google DeepMind | $260k - $380k | $450k - $650k | $750k - $1.2M |
| xAI | $250k - $400k | $450k - $700k | $700k - $1.5M |
The 6figr.com aggregator places OpenAI median TC at $612k, with the 90th percentile above $3.2M when the value of pre-IPO equity is marked to recent secondary prices. Anthropic engineers report base $180k to $300k plus RSU grants and 15 to 25% performance bonus, landing most at $300k to $490k.
If you are a founder and you read "ML engineer salary $250k" in a generic guide, the implicit assumption is FAANG production MLE in California. That's a real job, but it is not the job 95% of seed-stage startups are actually trying to hire for.
Global pay map: US, EU, India, LATAM, Asia
Region matters more for ML than for backend or frontend, because frontier-lab gravity has pulled US comp into a different zip code while applied MLE work is increasingly global.
| Region | Junior base | Mid base | Senior base | Notes |
|---|---|---|---|---|
| US (national) | $115k | $148k | $190k | Built In national avg base $148k; SF +21% |
| UK | £45k (~$57k) | £65k (~$82k) | £95k (~$120k) | London median £67k (Glassdoor) |
| Germany | €58k | €82k | €115k | Strong Industrial AI freelance market at €90-130/hr |
| Switzerland | CHF 100k | CHF 130k | CHF 155k | Highest in Europe (~$145k senior base) |
| Poland | $25k | $40k | $53k | Largest CEE talent pool |
| Czech Republic | $36k | $44k | $54k | |
| Ukraine | $24k | $42k | $65k | War-time discount, strong talent |
| India | ₹6-10L | ₹12-20L | ₹25-50L | ($7k-$60k); top remote senior to US firms hits $80k-$130k |
| Brazil | $18k | $32k | $58k | Time-zone aligned to US |
| Mexico | $22k | $38k | $62k | |
| Argentina | $15k | $28k | $50k | |
| Singapore | S$80k | S$130k | S$185k | (~$140k senior USD) |
Sources: Glassdoor 2026 country pages, Levels.fyi, Optiveum 2025-2026 country guide, Alcor LATAM benchmarks, igmGuru India tracker.
The headline gap: a senior MLE in Eastern Europe or LATAM costs roughly half a US senior, and a senior in India costs a third. With remote tooling and async ML pipelines being the norm in 2026, the productivity penalty has mostly evaporated for applied work. Frontier research is the exception; that work still concentrates in three US cities and two London zip codes.
What the salary table doesn't tell you
Every comp guide stops at base + equity. The fully-loaded cost of an ML engineer hire in the US is closer to 1.6x to 1.8x base, before you account for the engineer not actually shipping for the first quarter.
- Benefits load: 25 to 30% of base in the US (health, payroll tax, 401k match, FSA).
- Recruiter fee: 20 to 25% of first-year base for an external retained search. ML talent is contested, so contingency rates run higher than backend.
- Equipment, GPU credits, tooling: $400 to $1,500 per month per engineer. Cursor, Claude, Copilot, GitHub, Notion, Linear, vector DB seats, plus a Modal or RunPod budget if they need GPUs.
- Time to first shipped model: 8 to 14 weeks for an applied MLE, 4 to 6 months for production ranking work. You pay the same salary while the engineer onboards.
- Turnover risk: ML engineer attrition in the US is around 22% in year one, higher in early-stage startups where the role is still being scoped. Each replacement resets the 8 to 14 week ramp.
Run the math on a $190k senior MLE base. Fully loaded with benefits, recruiter, equipment, and a 12-week ramp where they ship roughly 30% of normal output, the real first-year cost lands closer to $340k for ~70% of one engineer's productive output. That is the number to compare against alternatives, not the $190k headline.
For a deeper version of the fully-loaded cost spreadsheet across roles, our senior software engineer salary by region 2026 breakdown has the same math applied to generalist SWE comp.
The on-demand alternative
Weekly booking changes the unit economics for any MLE role that is not a 5-year strategic capability. Cadence pricing is locked at four tiers:
- Junior, $500/week for cleanup, integrations with good docs, eval-set wiring, prompt scaffolding.
- Mid, $1,000/week for standard model integration work, fine-tuning a Llama or Mistral variant, building RAG pipelines, end-to-end shipping.
- Senior, $1,500/week for owning ML scope, designing eval frameworks, MLOps architecture, complex refactors.
- Lead, $2,000/week for fractional ML CTO work, full architecture, GPU-cluster decisions, vendor calls.
Annualized, a senior on Cadence is $78,000 for 52 weeks of work. The same senior in-house at $190k base + 30% benefits + recruiter amortization runs $285k to $340k fully loaded. That is roughly 3.6x to 4.4x more expensive for the same scope of work.
The math, side by side:
| Approach | Annual cost (senior MLE) | Time to first commit | Replace in week | Long-term retention |
|---|---|---|---|---|
| US in-house hire | $285k - $340k loaded | 8 to 14 weeks | No (notice + severance) | Strong if scope fits |
| US contractor at $200/hr | $300k - $400k | 1 to 2 weeks | Yes (notice clause) | Weak |
| Toptal MLE | ~$200k+ at $150/hr | 1 to 2 weeks | Notice required | Mid |
| Offshore agency | $80k - $140k | 3 to 6 weeks | Contract bound | Mid |
| Cadence senior weekly | $78k (52 wks) | 27-hour median to first commit | Yes, any week, no notice | Founder dependent |
Cadence is not always the right answer. If you are building a capability you will operate for 5+ years (your core ranking model, your fraud system, your moat), you want headcount with equity skin in the game. The booking model wins when you have a 4 to 40 week ML scope: ship the eval pipeline, fine-tune on your domain, build the RAG layer, integrate the agent. If you want to model your own scenario, our engineering ROI tool lets you compare in-house, contractor, and weekly booking on your actual numbers.
Why the 2026 market looks different from 2024
Three forces have rearranged the comp landscape in two years:
- Frontier-lab gravity well. OpenAI, Anthropic, xAI, and Google DeepMind have absorbed the top 1% of ML talent at $500k to $3M+ packages. This pulled FAANG production MLE comp up roughly 18 to 25% to defend (Levels.fyi 2024 to 2026 deltas).
- Applied MLE became a crowded job. Cursor, Claude Code, and Copilot let backend engineers ship LLM features without the ML PhD. Many companies discovered they did not need an "ML engineer" at all; they needed a senior backend engineer who could write good prompts and wire eval sets. This is core to why every engineer on Cadence is AI-native by default, vetted on Cursor and Claude Code fluency before they unlock bookings. The line between "ML engineer" and "AI-native backend" has blurred.
- Global supply caught up on applied work. Senior applied MLEs in Bangalore, São Paulo, and Warsaw now ship production-grade RAG, fine-tuning, and eval pipelines indistinguishable from US output. Frontier research is still concentrated; everything downstream is not.
The practical takeaway: scope your role honestly before you write a comp band. If you need a fine-tune + a RAG layer + an eval harness, that is a 6 to 12 week mid-engineer scope at $6k to $12k total on Cadence, not a $250k headcount hire. If you need a recommendation system that 80% of your DAUs hit, that is a senior or staff MLE on payroll. Both can be right; one is right far more often than founders think.
How to use this data: a 5-question decision framework
Before you commit to a $250k+ MLE budget, answer these:
- Is this an ML problem or an LLM-API problem? GPT-4-class APIs, Claude, and Gemini handle 70% of "we need ML" requests in 2026 with no model training. The right answer might be a backend engineer with prompting fluency. For help deciding, our build-vs-buy-vs-book tool takes a feature spec and routes you.
- Is this a 12-week project or a 5-year capability? Project-shaped ML work is dramatically cheaper to book by the week. Capability-shaped work needs equity-aligned headcount.
- Do I have the eval infrastructure to know if the model is working? If no, the first hire is whoever builds that, and it is rarely a $400k research engineer.
- What is my replacement cost if this hire doesn't work? A bad senior MLE hire in the US costs roughly $180k to $250k (severance, recruiter re-fee, lost productivity, opportunity cost). On a weekly model that number is the cost of one bad week.
- Am I confusing "ML engineer" with "AI-native backend engineer"? Look at the actual scope. If the answer is "wire an LLM into our app and ship features", you want a senior backend who codes with Claude every day, not an MLE.
For a related lens on adjacent roles, see how data engineer comp and AI engineer comp have moved over the same period; the gaps explain a lot about where the market is heading.
If you have ML scope in the next 4 weeks, the fastest test is to book a senior on Cadence at $1,500 for the week and use the 48-hour free trial to see what they ship. If they do not move the needle on day one, you replace them; if they do, the year-one cost is a quarter of a US headcount hire. Run your own numbers on the Cadence ROI calculator before you write the job description.
Sources
- Levels.fyi 2026 company pages (Google, Meta, Amazon, Apple, LinkedIn, OpenAI, Anthropic, Snap)
- Glassdoor 2026 base salary medians (US, UK, Germany, India, Brazil)
- Built In 2026 ML engineer salary report
- 6figr.com OpenAI compensation aggregator
- Optiveum Machine Learning Engineer Salaries by Country 2025-2026
- Alcor AI Engineer Salary by Country 2026
- BLS Occupational Employment and Wage Statistics, computer and information research scientists
- igmGuru ML engineer India salary tracker 2026
- Stack Overflow Developer Survey 2025 (ML/AI specialization band)
FAQ
What does an ML engineer make in the US in 2026?
The US median base is roughly $148,000 and median total comp around $212,000 for a generalist production MLE. FAANG medians range from $290k (Google) to $430k (Meta), and frontier labs like OpenAI sit at a $612k median total compensation with senior packages above $1M.
How much do OpenAI and Anthropic ML engineers actually make?
OpenAI software and research engineers run $249k to $1.28M+ at L2 to L6, with a $795k median total compensation per Levels.fyi. Anthropic engineer total comp lands between $300k and $490k typical, with senior research scientists reaching $550k.
Is an ML engineer worth more than a backend engineer in 2026?
For genuine modeling work (training, evals, MLOps), yes, roughly 15 to 40% premium over a same-level backend engineer. For LLM-API integration work, the premium has nearly disappeared because every Cadence engineer (and most senior backend engineers in 2026) ship LLM features as a baseline skill.
What does an ML engineer cost outside the US?
Senior ML engineer base salaries in 2026: UK ~£95k ($120k), Germany ~€115k, Poland ~$53k, India ~$30-60k (top remote seniors to US firms hit $80-130k), Brazil ~$58k. Eastern Europe and LATAM run roughly half US comp; India runs a third for similar applied work.
When should I hire an in-house MLE versus book one weekly?
Hire in-house when the scope is a 5-year strategic capability with equity-aligned ownership (your core ranking model, your moat). Book weekly when the scope is a 4 to 40 week project (eval pipeline, fine-tune, RAG layer, agent). The fully-loaded headcount cost of a US senior MLE is roughly 3.6x to 4.4x the annualized cost of a senior on Cadence at $1,500/week, so the math heavily favors booking for project-shaped work.