H-1B Lottery Odds by Wage Level (FY2027)
How the weighted H-1B draw works, realistic odds at Levels I to IV, the salary each level needs by metro, and where posted US AI pay lands.
For most of its history the H-1B lottery treated everyone the same. A new graduate on a starting salary and a staff engineer on a senior package each held one ticket. That changed for the FY2027 season. The draw held in March 2026 was the first to give more entries to higher-paid offers, and the number on your offer letter now shapes your chances almost as much as luck does.
The headline figures you may have seen, 15% at Level I and 61% at Level IV, come from the government's own estimate. They are useful, but they are a model built on an older and larger applicant pool, and they slightly overstate the advantage at the top. This piece explains how the weighted draw works, what the published odds really say, what they might look like with a smaller pool, and, using live postings in our index, which wage level US software, data and AI jobs would actually land at.
This reflects published rules and data as of 19 October 2026. It is a plain-language summary, not legal advice. Every odds figure below is an estimate, and each one shows how it was made.
How does the weighted H-1B lottery work?
A final rule published on 29 December 2025 took effect on 27 February 2026, in time for FY2027 registration. Each registration is placed in the selection pool a number of times based on the wage level of the offered job. The levels are the four Occupational Employment and Wage Statistics (OEWS) levels the Department of Labor publishes for every occupation and area.
| Wage level | DOL description | Entries in the pool |
|---|---|---|
| Level I | Entry | 1 |
| Level II | Qualified | 2 |
| Level III | Experienced | 3 |
| Level IV | Fully competent | 4 |
The employer picks the highest level that the offered wage equals or exceeds, for the job's occupation code and the area where the work happens. A few details matter more than they first appear:
- Entries are not selections. A Level IV person has four tickets in the drum, but can only be picked once and only counts once against the cap.
- Several locations means the lowest level. If the job covers more than one work area, the employer must use the lowest level the wage reaches across them.
- Several employers means the lowest level too. If two employers register you at different levels, USCIS uses the lowest. A second, cheaper offer does not add a ticket; it can take tickets away.
- Below Level I still counts as Level I. If the employer relies on a lower wage from another legitimate survey, the registration goes in at Level I. Nobody is shut out.
The $215 registration fee and the separate $100,000 payment are a different story, covered in our guide to the 2026 H-1B fee and your alternatives.
What happened in the FY2027 lottery?
Registration ran from 4 to 19 March 2026. On 17 July 2026 USCIS announced it had received enough petitions to fill both the 65,000 regular cap and the 20,000 US advanced-degree exemption. There will be no second FY2027 draw, so anyone not picked in March waits for the FY2028 registration in March 2027.
What USCIS has not yet published is the FY2027 row of its registration table. As of its 21 September 2026 update, the latest row is still FY2026. So there is no official count yet of how many people registered at each wage level, or how many were selected at each level. Any FY2027 odds you see quoted are estimates.
For context, the last season with full numbers, FY2026, had 336,153 eligible unique beneficiaries and 120,141 selected registrations.
One law firm, Mintz, reported about 211,600 unique beneficiary registrations for FY2027. That is a secondary, rounded figure, not a USCIS table, but if it holds, the pool shrank by roughly a third in a single year. A smaller pool changes the odds for everyone, which is why we model it below.
What are the H-1B lottery odds by wage level?
The final rule estimates selection chances using a five-year average (FY2020 to FY2024): about 320,711 eligible people a year, of whom 28% were at Level I, 55% at Level II, 12% at Level III and 5% at Level IV, with about 94,900 selections. It then reports two sets of numbers, and they are not the same.
| Wage level | Old equal-odds lottery | DHS simple estimate | DHS simulation | Our re-run of the same inputs |
|---|---|---|---|---|
| Level I | 29.6% | 15.3% | 17.0% | 17.0% |
| Level II | 29.6% | 30.6% | 31.1% | 31.1% |
| Level III | 29.6% | 45.9% | 42.8% | 42.9% |
| Level IV | 29.6% | 61.2% | 52.6% | 52.6% |
How the simple estimate works. DHS divides the selections by the total number of tickets: 94,900 divided by (89,911 x 1 + 177,216 x 2 + 37,928 x 3 + 15,657 x 4), which gives 15.29% for each ticket. Levels II to IV are then that figure times two, three and four.
Why the simulation is lower at the top. In a real draw, once a Level IV person is picked, their other three tickets leave the drum. The simple method ignores this, so it overstates the advantage of extra tickets. DHS also ran a simulation that removes winners as it goes, and its table shows 15,330 of 89,911 Level I registrants selected, about 17%, with the other levels shown in the same table. We rebuilt that draw ourselves (method at the end) and landed on the same figures to within a tenth of a point.
So the more realistic reading of the government's own model is roughly 17%, 31%, 43% and 53%. Level IV is about three times Level I, not four.
What might the odds look like with a smaller pool?
The model above assumes about 320,000 people. If FY2027 really had about 211,600, every level does better, because the same number of slots is shared among fewer people. The open question is how many registrations USCIS selects, since it always picks more than 85,000 to allow for people who never file. It selected about 120,000 registrations for FY2026.
Here is our simulation with a pool of 211,600 and the historical wage mix, across a range of selection counts:
| Selections made | Level I | Level II | Level III | Level IV |
|---|---|---|---|---|
| 85,000 (cap only, nobody drops out) | 24% | 43% | 56% | 67% |
| 100,000 | 29% | 50% | 65% | 75% |
| 120,000 (similar to FY2026) | 37% | 60% | 75% | 84% |
Two things push these numbers down. First, the wage mix probably moved: the people who stopped registering were most likely lower-wage cases. If the mix shifts to 20%, 50%, 20% and 10%, Level I falls by about two to four points in each row. Second, the 20,000 master's cap is a separate second stage, which helps people with a qualifying US master's degree and slightly hurts everyone else. DHS, like us, models one pooled draw for simplicity.
Treat this table as a range, not a forecast. The honest summary is that a Level I registration in a pool of this size probably had somewhere between a one-in-five and a one-in-three chance, and Level IV somewhere between three-in-five and four-in-five.
Odds also compound over repeat seasons, if your offer stays at the same level and the draw looks similar each year:
| Wage level (DHS simulation odds) | One season | Two seasons | Three seasons |
|---|---|---|---|
| Level I (17%) | 17% | 31% | 43% |
| Level II (31%) | 31% | 52% | 67% |
| Level III (43%) | 43% | 67% | 81% |
| Level IV (53%) | 53% | 78% | 89% |
One season at Level III is about the same as three seasons at Level I.
What salary puts you at each wage level?
The thresholds depend on the occupation code and the metro, and they reset every July. The table below uses DOL's wage year that runs from 1 July 2026 to 30 June 2027, which is the data in force for a March 2027 registration (FY2028). The FY2027 draw used the previous year's data, which was a few percent lower; for data scientists in San Francisco, Level I was $103,522 then and $110,386 now.
| Occupation | Metro | Level I | Level II | Level III | Level IV |
|---|---|---|---|---|---|
| Software developers (15-1252) | San Francisco | $137,093 | $165,131 | $193,149 | $221,187 |
| Software developers (15-1252) | New York | $109,845 | $137,717 | $165,589 | $193,461 |
| Software developers (15-1252) | Austin | $95,888 | $119,642 | $143,395 | $167,149 |
| Data scientists (15-2051) | San Francisco | $110,386 | $142,439 | $174,512 | $206,565 |
| Data scientists (15-2051) | Seattle | $93,912 | $127,026 | $160,119 | $193,232 |
| Data scientists (15-2051) | New York | $87,256 | $115,836 | $144,415 | $172,994 |
| Data scientists (15-2051) | Austin | $77,335 | $101,962 | $126,589 | $151,216 |
Metro names are shortened; each is the full DOL area (for example, San Francisco means San Francisco, Oakland and Fremont, and excludes San Jose and Silicon Valley, which have their own figures). We could not confirm the Seattle software developer figures, so they are left out. Check the official FLAG wage search for your exact occupation and county before relying on any of these.
There is no "machine learning engineer" occupation code. An ML engineer who mostly builds production software usually fits software developers; one who mostly models data may fit data scientists; research roles may fit computer and information research scientists, which has its own table. The code follows the duties, not the title, and the employer chooses it. Since data scientist thresholds are lower in every metro, the same salary can land a level higher under that code.
Which wage level do posted US software and AI salaries land at?
This is where our own data comes in. We took live US postings in our index that list an annual base pay range and are located in one of these metros, then checked which level the bottom of the range would clear. The floor is a cautious stand-in for the offer; the actual offer is often higher.
| Role and metro (threshold used) | Below Level I | Level I | Level II | Level III | Level IV | Level III or above |
|---|---|---|---|---|---|---|
| Software, San Francisco | 13% | 22% | 22% | 18% | 25% | 43% |
| Software, New York | 6% | 11% | 26% | 26% | 31% | 57% |
| Software, Austin | 11% | 24% | 24% | 14% | 27% | 41% |
| AI and ML, San Francisco (software developer) | 6% | 19% | 19% | 24% | 32% | 56% |
| AI and ML, New York (software developer) | 4% | 7% | 22% | 24% | 43% | 67% |
| AI and ML, Seattle (data scientist) | 1% | 13% | 36% | 25% | 24% | 49% |
| Data science, San Francisco | 8% | 16% | 22% | 24% | 29% | 53% |
| Data science, New York | 10% | 14% | 27% | 20% | 29% | 49% |
The pattern is clear. The typical posted AI role would register at Level III or IV, even on the bottom of its range. The median AI floor in San Francisco is $200,000, which clears Level III for software developers. On the midpoint of the range, about two in three AI postings in San Francisco and New York clear Level IV.
General software is more split. The median San Francisco floor of $180,000 sits at Level II, while New York's $175,000 already clears Level III because New York's thresholds are lower. Data science sits between the two.
Seniority explains most of the spread. Pooling software postings across San Francisco, New York and Austin, each measured against its own metro:
| Seniority | Below Level I | Level I | Level II | Level III | Level IV |
|---|---|---|---|---|---|
| Entry, software | 34% | 35% | 20% | 6% | 5% |
| Mid, software | 21% | 25% | 27% | 15% | 12% |
| Senior, software | 7% | 17% | 27% | 27% | 22% |
| Staff, software | 4% | 10% | 17% | 24% | 45% |
| Mid, AI and ML | 13% | 26% | 26% | 24% | 11% |
| Senior, AI and ML | 2% | 12% | 27% | 30% | 29% |
| Staff, AI and ML | 0% | 5% | 9% | 25% | 61% |
Nearly seven in ten entry-level software floors sit at or below Level I. A floor below the Level I figure does not mean the job cannot be sponsored, but such a registration would go in with one ticket. This is the rule's design showing up in real postings: it is the junior hire who loses ground. For pay benchmarks by seniority, see our AI engineer and ML engineer salary pages.
Do the companies posting these jobs sponsor H-1B visas?
Most postings do not say. Across live US postings in our index:
| Role family | Says it sponsors | Says it does not | Says nothing | Of those that say anything, share saying yes |
|---|---|---|---|---|
| AI and machine learning | 2.5% | 6% | 91% | 28% |
| Research science | 6% | 13% | 80% | 29% |
| Software engineering | 2.6% | 13% | 84% | 16% |
| Data science and analytics | 1.2% | 12% | 86% | 9% |
| Data engineering | 1.2% | 12% | 87% | 9% |
Silence is not a no. We matched the companies behind these postings to USCIS's H-1B Employer Data Hub. About 61% of silent US AI postings, 59% of silent software postings and 68% of silent data science postings come from employers with at least one H-1B approval in FY2025 or FY2026 to date. That is a lower bound, since we only counted confident name matches. Even about half of AI postings that say "no sponsorship" are at employers that sponsored someone, which suggests the answer is often about the specific role.
Our H-1B sponsorship jobs page lists US roles whose posting says the company can sponsor, and the wider visa sponsorship hub covers other countries.
What should you do with this?
- Ask which wage level the employer will register at. It belongs in any offer comparison next to base pay. On DHS's own simulation, a Level III offer gives about two and a half times the one-season chance of a Level I offer.
- Ask about the occupation code. For ML and data roles, the code chosen can move you a full level at the same salary. It must match your actual duties, so this is a question, not a negotiation trick.
- Watch the July wage reset. Thresholds rose between the FY2027 and FY2028 data. An offer that cleared a level last year may fall short next March.
- Do not read silence as a no. Most AI postings say nothing, and most of those employers have sponsored before. Ask early.
How we got these numbers (19 October 2026)
Odds. We modelled the draw as sequential weighted sampling without replacement: each person gets 1 to 4 tickets, a ticket is drawn, its owner is selected and their other tickets removed, until the selection count is reached. We ran it 20 times per scenario and averaged. With DHS's inputs (320,711 people, the 28/55/12/5 mix, 94,900 selections) it reproduces DHS's own published simulation. The smaller-pool scenarios use Mintz's approximate 211,600 registrations, which USCIS has not confirmed, and selection counts we chose to bracket plausible outcomes. They pool the regular and master's caps into one draw. Multi-season figures assume the same odds each year and independent draws.
Wage thresholds. DOL OFLC wage data for 1 July 2026 to 30 June 2027, read from a third-party reproduction of the DOL files. The equal spacing between levels in every row matches how DOL builds them, but we did not check each figure against the FLAG search directly.
Posted pay. Live, non-duplicate US postings in the landed index with an annual USD pay range, excluding on-target-earnings ranges, where every US city on the posting falls in one metro. We used our role family and seniority classification. San Francisco means the San Francisco, Oakland and Fremont area. About two thirds of software postings in San Francisco and New York list pay; only about a quarter in Austin do, since Texas has no pay transparency law, so the Austin figures lean toward employers that publish. AI postings in Austin and Seattle software postings were left out (too few postings, or no confirmed threshold). Posted ranges are base pay, not the employer's registered wage.
Sponsorship. Each posting's sponsorship position is read from its text. Employer approvals come from the USCIS H-1B Employer Data Hub (FY2025 and FY2026 through the third quarter), matched to the companies in our index, counting only confident matches.
Sources
- Federal Register: Weighted Selection Process for Registrants and Petitioners Seeking To File Cap-Subject H-1B Petitions, final rule, 90 FR 60864 (29 December 2025), including Table 13 and its regulatory impact analysis.
- USCIS: H-1B Weighted Selection Small Entity Compliance Guide (12 January 2026).
- USCIS: H-1B Electronic Registration Process (updated 21 September 2026).
- USCIS: USCIS Reaches Fiscal Year 2027 H-1B Cap (17 July 2026).
- USCIS: H-1B Employer Data Hub, FY2025 and FY2026 through Q3.
- US Department of Labor, Office of Foreign Labor Certification: FLAG wage data downloads and wage search, July 2026 to June 2027 wage year (updated 1 July 2026).
- US Department of Labor: Prevailing Wage Determination Policy Guidance, Nonagricultural Immigration Programs (revised November 2009).
- O*NET OnLine: Software Developers (15-1252), Data Scientists (15-2051), Computer and Information Research Scientists (15-1221).
- Mintz: USCIS Announces FY2027 H-1B Cap Reached, No Second Lottery to Be Conducted (20 July 2026).
- H1BGrader: reproductions of DOL OFLC prevailing wage levels, 2026 to 2027, for the metros listed.
- landed job index, live US postings on 19 October 2026.
Frequently asked
What are the H-1B lottery odds by wage level?
DHS's simple estimate is about 15%, 31%, 46% and 61% for Levels I to IV. Its own simulation, which removes people once picked, gives about 17%, 31%, 43% and 53%. Both use an older pool of about 320,000 people, so they are estimates, not FY2027 results.
How many entries does each wage level get in the H-1B lottery?
Level I gets one entry, Level II two, Level III three and Level IV four. Each person can still only be selected once and counts once against the cap.
Has USCIS published FY2027 selection numbers by wage level?
Not as of 19 October 2026. USCIS confirmed on 17 July 2026 that both FY2027 caps were reached and there is no second draw, but its registration table still ends at FY2026.
What salary do I need for Level IV in San Francisco?
For software developers in the San Francisco, Oakland and Fremont area, DOL's July 2026 to June 2027 data puts Level IV at about $221,187 a year. Data scientists need about $206,565. Check the FLAG wage search for your exact occupation and county.
What if two employers register me at different wage levels?
USCIS uses the lowest level among your registrations, so a second offer at a lower wage can reduce your odds rather than add a ticket. This is a summary of the rule, not legal advice.
Done reading? Let landed apply.
landed finds roles like the ones in this piece, applies on each company's own site in the first hours, and writes every answer. Your first 10 applications are free.
Get 5x interviews