Compensation intelligence · Up Top Search
What AI and crypto employers actually post.
Source-linked pay ranges from company job boards, Up Top’s compensation model kept separate and labelled, and package math that never blends base, bonus, equity, and token value.
- Postings in market
- 10,138
- With a USD range
- 5,226
- Companies
- 161
- Verified open
- Nov 4, 2026
Employer-posted base salary only. Equity, bonus, and token value are modelled separately and say so.
Observed · employer postings
What employers are posting right now.
Base-salary ranges from 161 companies’ own job boards, verified open on 2026-11-04. Filter, read the spread, and open the posting. Percentiles appear only above the display rule; otherwise you see examples, not a distribution.
AI. AI-native companies: frontier labs, AI model developers, AI infrastructure & compute, AI hardware & silicon, applied-AI startups, and physical AI & robotics. Membership requires a reviewed cohort at medium or high confidence.
153 companies · 10,046 open postings · 5,174 with a USD annual range · cohorts: Frontier labs, AI model developers, AI infrastructure & compute, AI hardware & silicon, Applied-AI startups, Physical AI & robotics
Loading 10,138 observations and verifying the release checksum…
Observed · by employer
Who is setting the range.
Per-company medians for the AI market, with each employer’s share of the postings that disclose pay. Compare the posting-weighted midpoint with the company-balanced one before quoting a market number.
Waiting for the verified release…
Arithmetic · your inputs
Compare two packages without blending them.
Base, target bonus, commission, sign-on, equity vesting, and token vesting stay on separate lines. First-year and recurring totals are computed at grant value; the risk-adjusted lines apply Up Top’s discount assumptions and are labelled as such.
| Component | Package A | Package B | B − A |
|---|---|---|---|
| Base salary | $255,000 | $230,000 | −$25,000 |
| Target bonusbase × target % | $25,500 | $0 | −$25,500 |
| Commission / variable (annual) | $0 | $0 | — |
| Sign-on (year one only) | $0 | $25,000 | +$25,000 |
| Equity vesting — year onegrant ÷ vest years; zero if the cliff is longer than 12 months | $100,000 | $150,000 | +$50,000 |
| Equity vesting — recurring annual | $100,000 | $150,000 | +$50,000 |
| Target cash (annual) | $280,500 | $230,000 | −$50,500 |
| First-year totalcash + sign-on + year-one vesting at grant value | $380,500 | $405,000 | +$24,500 |
| Recurring annual totalcash + annual vesting at grant value | $380,500 | $380,000 | −$500 |
| Risk-adjusted equity (annual)model assumptionUp Top assumption: 45.5% retained after Vesting discount 35% · Liquidity discount 30% | $45,500 | $68,250 | +$22,750 |
| Risk-adjusted recurring totalmodel assumption | $326,000 | $298,250 | −$27,750 |
Equity and token lines use grant-date value spread evenly over the vest; real vesting schedules, refreshes, strike prices, and tax are not modelled. Risk-adjusted lines apply Up Top’s discount stack as haircuts to retained value; they are assumptions, not market data, and sit below the totals on purpose.
Modelled · Up Top assumptions
Up Top’s compensation model.
Band midpoint × level × geography × stage × scarcity gives an adjusted base salary. These are the assumptions Up Top uses in search work — not observed data, and not to be layered on the posted ranges above.
The score multiplier is how Up Top prices candidate quality internally. It is not an observed market relationship and has not been validated against outcomes.
Adjusted base (Up Top model)
$223,250
AI infra / GPU engineer · Senior · North America (US / Canada) · Series A
- Band midpoint
- $235,000
- Band range
- $200,000–$270,000
- Level
- 1.00×
- Geography
- 1.00×
- Stage
- 0.95×
- Scarcity
- 1.00×
Adjusted base = band midpoint × level × geography × stage × scarcity. Base salary only. Do not apply these multipliers to a posted range above — postings already reflect the employer’s geography and stage.
Band matrix — AI model 1.1.0
Up Top reference bands (USD, North America baseline) assembled from levels.fyi, Carta's 2026 AI compensation analysis, and the Perspective AI FDE survey. Assumptions — compare with the posted ranges above. Click a row to load it into the calculator.
| Category | Junior | Mid | Senior | Lead / Staff | Executive |
|---|---|---|---|---|---|
| $130K$110K–$150K | $190K$160K–$220K | $243K$220K–$270K | $320K$280K–$360K | $400K$340K–$450K | |
| $135K$115K–$155K | $185K$150K–$220K | $235K$200K–$270K | $300K$260K–$340K | $370K$320K–$420K | |
| $150K$150K | $203K$180K–$225K | $255K$225K–$255K | $290K$260K–$320K | $340K$300K–$380K | |
| $130K$110K–$150K | $170K$150K–$190K | $200K$180K–$220K | $235K$210K–$260K | $270K$240K–$300K | |
| $95K$80K–$110K | $145K$120K–$170K | $205K$170K–$240K | $245K$200K–$290K | $260K$210K–$310K | |
| $70K$55K–$85K | $120K$95K–$145K | $175K$145K–$205K | $205K$165K–$245K | $235K$185K–$285K | |
| $150K$125K–$175K | $190K$155K–$225K | $225K$185K–$265K | $240K$195K–$285K | $255K$205K–$305K | |
| Forward deployed engineertotal comp · reference onlyPerspective AI — State of Forward Deployed Engineering 2026 | $200K$170K–$230K | $300K$250K–$350K | $385K$340K–$430K | $610K$550K–$670K | $1.1M$1M–$1.2M |
Forward deployed engineer: total compensation, not base.
Survey of 1,500 FDEs; compensation reported as median annual cash + target bonus + estimated equity at the last preferred-share price (US). The senior $385K is the survey's frontier-lab figure after its own 40% tender-equity discount; the headline is $485K. Total compensation, not base.
Equity allocation by seniority
- Junior
- 0.01% – 0.05% of supply
- Mid
- 0.05% – 0.146% of supply
- Senior
- 0.1% – 0.25% of supply
- Lead / Staff
- 0.2% – 0.4% of supply
- Executive
- 0.5% – 1% of supply
Discount stack Up Top model assumption
- Vesting discount4-year vest with 1-year cliff; illiquid during lockup · range 30%–40%
- 35% haircut → 65% kept
- Liquidity discountPrivate equity — infrequent tenders; 409A can lag · range 20%–40%
- 30% haircut → 70% kept
- Retained share of grant value
- 45.5%
RSUs / options, not tokens. Retained share after the two haircuts is 0.65 × 0.70 = 45.5%, consistent with the ~45–65% range in the source file.
Method
How to read these numbers.
Sources, definitions, exclusions, the display rule, and the release’s own quality checks.
Two kinds of numbers
Observed: ranges employers published on their own applicant-tracking boards (Ashby, Greenhouse, Lever, Google Careers), pulled first-party by Up Top’s monitor and released as a versioned dataset. Every observation links to its posting. Modelled: Up Top’s band matrix, multipliers, score tiers, and discount stacks — assumptions we use in search work, labelled wherever they appear. The two are never mixed in one figure.
What a posted range is
base_salary as posted. Bonus, commission/OTE, sign-on, equity, and token value are not in these numbers; equity_offered is the board's boolean only.
medium on Sales/BD and Solutions families because a posted band there may be an on-target figure the source does not label.
Statistics are named for what they measure: percentiles of posted lower bounds, of posted upper bounds, and of range midpoints. A median upper bound is the median advertised ceiling — it is not median pay. Percentiles use only USD, annual, full-time-eligible ranges; counts and disclosure use every matched posting.
Display rule
Cohort percentiles appear only with at least 20 postings with pay from 5 employers. Below that the view broadens one dimension at a time and says so, or shows source-linked examples with no distribution claim. This is a product heuristic, set centrally in the release, not a statistical guarantee. Company-balanced medians and the employer concentration index are shown so a few large boards cannot quietly define a market.
What is not here
This exporter reads only the ATS-derived store and static config. Recruiter calls, candidate expectations, client budgets, offers, and accepted packages have no path into this artifact.
Bonus, commission, sign-on, equity, and token value are not in posted ranges. Funding stage is never inferred from a company name. Level comes from the title alone, so “level not stated” is common. Non-USD ranges (326), monthly or hourly rates (21), and intern, contract, or part-time postings (205) are kept as observations but excluded from every percentile. 19 exact duplicate postings were collapsed.
Model provenance
Crypto model 2026.02 (Feb 27, 2026): Up Top model bands, North America baseline, base salary in USD before multipliers. Calibrated from Up Top placements and search work; not a posting sample. AI model 1.1.0 (Sep 11, 2026): Up Top reference bands (USD, North America baseline) assembled from levels.fyi, Carta's 2026 AI compensation analysis, and the Perspective AI FDE survey. Assumptions — compare with the posted ranges above. The forward-deployed engineer row is a total-compensation survey figure and is shown as a reference, never as a base band; an earlier version of this file ran it through the base formula and then added equity, which double-counted.
- Release
- 2026-11-04
- Generated
- Nov 4, 2026
- Source batch
- 245/245 boards · complete
- Verified open on
- Nov 4, 2026
- Posted between
- Feb 8, 2016 – Nov 3, 2026
- Boards configured
- 245
- Open roles in store
- 12,781
- In a market
- 10,138 postings · 161 companies
- With USD annual range
- 5,226
- Schema / rules
- 1.0.0 / 2026-11-03.1
- Cohort review
- Sep 11, 2026
- Private evidence
- none
Company cohorts in this release
- Frontier labs
- Labs training general-purpose frontier foundation models.
- 7
- AI model developers
- Companies training their own domain or media models (voice, video, music, image, code, science, world models) that are not general frontier labs.
- 26
- AI infrastructure & compute
- GPU clouds and neoclouds, inference and serving platforms, vector databases, agent frameworks, evals and observability, and AI training-data vendors.
- 38
- AI hardware & silicon
- Companies designing AI accelerators and AI-specific compute hardware.
- 6
- Applied-AI startups
- AI-native application companies selling an AI product into a vertical or function.
- 67
- Physical AI & robotics
- Humanoids, robot foundation models, and autonomous vehicles. Boards carry many mechanical, electrical, and manufacturing roles.
- 9
- Crypto
- Crypto-native protocols, exchanges, market makers, wallets, and tokenization companies.
- 8
Excluded from every market: company not in any market 2,285 · company not in labs config or cohorts 18 · not confirmed in source batch 321.
Release quality checks
- passstore integrity okPRAGMA integrity_check = ok
- passlatest batch complete2026-11-04T00:01:22.457772+00:00 status=complete boards=245/245
- passno failed board in a marketno boards failed
- passevery board has cohortlabs without cohort: none
- passno duplicate cohort rowsduplicates: none
- passcohort rows match boardscohort rows without a board: none
- passorphan store companies excluded3 store companies not in current config, excluded: ['Cycles', 'Tempo', 'Vibiz']
- passno reversed ranges0 ranges arrived reversed and were swapped
- passno missing lower endpoint0 rows had an upper bound without a lower bound
- passnon usd excluded from usd stats326 non-USD posted ranges retained as observations, excluded from USD statistics
- passnon annual pay periods excluded21 non-annual ranges excluded from annual statistics
- passannual floor ceiling enforced6 below $40,000 · 0 above $2,000,000
- passnon full time excluded205 intern / part-time / contract / fellowship postings excluded from statistics
- passall evidence is employer postingno private evidence kinds are reachable from this exporter
- passevery observation has source url0 observations lack a public source URL
- passno market company below confidencemarket membership requires medium/high cohort confidence
- passstale open rows excluded321 rows marked open but not confirmed in the source batch were excluded
- passduplicates collapsed19 exact duplicate postings collapsed
- passhas observations10138 observations
No failing checks. 0 informational warning(s).
Refresh: run scripts/frontier_public_export.py in uptop-crons against the monitor store and commit the new public/data/ai-comp/ release. The exporter refuses any batch where a board in a market failed, and any failing check, so a broken collection leaves this page on the last valid release. A failed board in no market can’t change these numbers, so it appears above as a warning.