Pay is high across the board, since even the lowest-paid family sits at $249k. The technical roles cluster at the top, and AI research & ML leads at $370k, about 1.5 times that. Rarity alone does not explain the ladder, because safety work, scarcer still at 7% of openings, is paid less. And the companies hire most in go-to-market (29% of openings), not in the best-paid work. Only four of the six disclose their ranges, so the picture below is a US one.
Even the lowest-paid family clears $249k, while research at the top reaches $370k.
Among the four that disclose, Anthropic pays the most, because its median reaches $360k and 39% of its roles pay above $400k. OpenAI follows at $307k, then come Cohere at $225k and xAI at $201k. xAI's pay is split in two, however, since 47% of its disclosed roles fall below $200k while 38% cluster between $300k and $350k. This gap reflects its two workforces, namely lower-paid annotators and highly paid engineers.
How to read. The family table shows each of the six job families across all 1769 roles: its share of the market (bar on a 0 to 100% scale), its headcount, and its median pay. The company panel then shows, for each disclosing company, how its salaried roles spread across the pay bands, with the median at right. Coverage: only four publish pay, OpenAI (79% of roles), Anthropic (68%), xAI (40%) and Cohere (20%), largely because of US pay-transparency laws, so these figures describe the US market only, in USD; Mistral and ElevenLabs disclose nothing and appear as n.a. Pay is the midpoint of the posted range. Source: the six companies' career sites, July 2026.