Research
All 623 of these AI jobs state a salary. Some state it 3× more precisely than others.
Read off 49 companies’ own job boards on 28 August 2026. Every posting counted here publishes a figure — so this is not about who discloses. It is about what the disclosure is worth.
- median band, as a share of the midpoint
- 31.5%
- tightest — Webflow
- 18.2%
- widest — Figma
- 55%
“The salary is disclosed” is not one claim
Pay-transparency law asks employers to publish a range. It does not say how wide, and that turns out to be where all the variation lives. Across 623 annual ranges the median band is 31.5% of the midpoint — roughly $68k.
But the median hides the point. Webflow states a median band of $36k on a $213k midpoint. Figma states $132k on $236k. Both comply. One of them has told you roughly what the job pays.
This is the third of three measurements on the same harvest. The first counts who states a figure at all; the second reads what they stated. This one asks what the statement is worth.
Band width by employer, tightest first
15 employers published at least 8 priced roles. Ranked on band width as a share of the midpoint, because a $60k band on a $400k role is a tighter statement than $50k on $120k.
| Company | Priced roles | Median band | Share of midpoint |
|---|---|---|---|
| Webflow | 14 roles | $36k on $213k | 18.2% |
| Crusoe | 8 roles | $30k on $160k | 19.2% |
| Anthropic | 42 roles | $80k on $363k | 21.5% |
| Cognition | 13 roles | $40k on $200k | 22.2% |
| ClickHouse | 8 roles | $45k on $176k | 22.7% |
| Temporal | 44 roles | $55k on $245k | 23.1% |
| LangChain | 72 roles | $50k on $208k | 23.4% |
| Snowflake | 8 roles | $51k on $188k | 27% |
| Databricks | 37 roles | $68k on $214k | 31.6% |
| Vercel | 56 roles | $70k on $219k | 31.9% |
| Replit | 59 roles | $75k on $240k | 33.3% |
| OpenAI | 60 roles | $92k on $312k | 34.1% |
| LlamaIndex | 10 roles | $78k on $190k | 38.9% |
| Harvey | 9 roles | $68k on $179k | 40% |
| Figma | 91 roles | $132k on $236k | 55% |
A wide band is not the same as a dishonest one
The most common reason a range is wide is that one advert covers several levels — a junior-through-staff posting genuinely spans that much, and writing one advert instead of four is ordinary practice. The page does not read a wide band as evasion, and neither should you.
What it does mean is that the number resolves less. A band at 55% of the midpoint leaves about $132k unanswered, which for most readers is the whole question. 6 postings in the whole set quoted a single figure rather than a range.
Method
Every open role on 49 AI companies’ public applicant tracking systems was fetched on 28 August 2026 and its stated pay range read. Only annual ranges quoted in US dollars are counted: folding an hourly rate or a second currency into a width produces a figure no employer ever quoted.
An employer needs 8 priced roles to appear in the table. Below that a median describes a handful of postings rather than a company, so those roles are counted in the overall figure and left out of the ranking — the same threshold the other two pages use.
The figures come from a committed snapshot rather than a live query, so they move only when the harvest is re-run. A research page whose headline changed quietly between two readings would be worth less than none.
Questions
- Which AI companies publish the most precise salary ranges?
- Webflow. Its median posting states a band of $36k on a $213k midpoint — 18.2% of the number. Crusoe and Anthropic follow. Only employers with at least 8 priced roles are ranked, because a median over a handful of postings is a fact about those postings.
- How wide is a typical AI salary range?
- 31.5% of the midpoint, or about $68k, across 623 annual ranges. But that median hides the spread that matters: Figma's median band is 55% of its midpoint against Webflow's 18.2%, so two employers can both "disclose" and tell you very different amounts.
- Does a wide salary range mean a company is being evasive?
- Not necessarily, and the page does not claim it. A wide band usually means one advert covers several levels — a genuine and common practice. What it does mean is that the number tells the reader less: at 55% of the midpoint, Figma's median posting leaves roughly $132k unresolved.
- How was this measured?
- Every open role on these companies' public applicant tracking systems was fetched and its stated range read. Only annual ranges quoted in US dollars count, because mixing an hourly rate or a second currency into a width produces a figure nobody quoted. 623 ranges from 49 employers cleared that filter; 15 employers had the 8 priced roles needed to be ranked.
- Are these figures estimates?
- No. Every floor and ceiling is a number the employer published on its own job posting. Nothing is modelled, extrapolated or self-reported.
Every listing on this board publishes a real pay figure, because it is the one rule the board has.