Inference engineer salary in 2026: posted bands, levels and what they leave out

Every figure below is a base salary range that an employer published on its own job posting, read from the posting on 16 October 2026, unless it is labelled as an aggregator figure. Pay-transparency laws in California, New York, Washington and Colorado make US postings the best public data for this role. Two cautions apply to all of them. First, equity is separate and at frontier labs it can be as large as base or larger, so a base band is not a total package. Second, one requisition often spans several levels, which is why some bands are very wide: the top of a $280,000 to $850,000 band is not a typical offer, and the midpoint is not a meaningful number.

Titles are copied from the postings. A role is included when its posting is about serving, inference performance, kernels or the inference engine, not merely about machine learning.

Anthropic posts the widest bands, because its performance roles span many levels: Performance Engineer, Inference Systems and Performance Engineer, Inference Engine at $350,000 to $850,000; Performance Engineer, GPU and TPU Kernel Engineer at $280,000 to $850,000; Staff and Senior Software Engineer, Inference Infrastructure at $320,000 to $485,000; and Engineering Manager, Inference Infrastructure at $405,000 to $625,000. Anthropic labels these ranges "Annual Salary".

OpenAI posts Software Engineer, Inference - Performance Optimization at $266,000 to $500,000, Software Engineer, Model Runtime and Software Engineer, Kernel Performance & AI Tooling at $266,000 to $445,000, and Software Engineer, Inference - Multi Modal at $295,000 to $555,000. xAI posts Software Engineer - Kernels/CUDA (C++) at $180,000 to $440,000.

NVIDIA states its base range per internal level, and the ladder is consistent across its inference postings: Level 2 $124,000 to $195,500, Level 3 $152,000 to $241,500, Level 4 $184,000 to $287,500, Level 5 $224,000 to $356,500 and Level 6 $272,000 to $431,250. Examples include Senior Software Engineer - AI Inference Performance (Levels 4 and 5), Senior Inference Engineer, GPU Kernel Optimization (Level 4) and a new-graduate TensorRT performance role at Levels 2 and 3. These are NVIDIA's own level numbers; they do not map one-to-one onto other companies' levels or onto aggregator level labels.

Perplexity posts Member of Technical Staff (AI Inference Engineer) at $220,000 to $485,000. Baseten posts both Software Engineer - Inference Performance and Software Engineer - GPU Kernels at $180,000 to $360,000. Modal posts Member of Technical Staff - Inference Runtime at $220,000 to $300,000. Together AI posts its inference and compute infrastructure engineering roles, junior through staff, at $160,000 to $280,000. CoreWeave posts Staff Software Engineer, Inference at $188,000 to $275,000 and Senior Software Engineer - GPU Kernel Authoring & Optimization at $182,000 to $242,000.

Larger companies post narrower, single-level bands: Databricks lists Staff Software Engineer - GenAI inference at $190,900 to $232,800 for San Francisco, and AWS lists Software Engineer-AI/ML, Inference Team - AWS Neuron at $143,700 to $194,400 for Seattle. Cohere publishes tiers by geography for its model efficiency roles: $205,000 to $380,000 in California, New York and Washington, and $175,000 to $325,000 elsewhere in the US.

Self-reported totals fill in what postings omit. On Levels.fyi in October 2026, the median total compensation for software engineers was $882,500 at Anthropic (49 submissions), $875,000 at OpenAI (240) and $312,000 at NVIDIA (2,345), against $250,000 for the site's ML and AI software engineer tag across all companies. These cover all software engineers at each company, not inference specifically, and are self-reported, so treat them as an order of magnitude for total pay at senior levels rather than as an inference benchmark.

Public posted bands outside the US are rare. Anthropic posts Staff Software Engineer, Inference at £325,000 to £390,000 in London and €295,000 to €355,000 in Dublin, and Senior Software Engineer, Inference at €235,000 to €295,000 in Dublin. None of the other employer boards checked published a range for London, Paris, Amsterdam or Zurich inference roles, and none published a rupee range for an Indian inference or kernel role.

Aggregator medians for the generic machine learning engineer title, total compensation, all companies, are about £112,000 in London, CHF 151,000 in Zurich, 44 lakh rupees in Bengaluru and 30 lakh in Hyderabad. They are not inference figures: the one like-for-like comparison, Anthropic's London staff inference band, is about three times the London median in base alone. Specialist inference roles at labs and accelerator companies in these cities sit well above the generic medians, and offers vary far more than the medians suggest.

Level does most of the work, and level is decided by the loop: the depth of the system design and debugging rounds, and whether your answers carry numbers. Specialization is the second lever: kernel and performance roles at labs carry the widest bands in this data. Location still matters within the US, as Cohere's tiers show, though less at remote-friendly employers.