The Upwork market in Q2 2026: 74,000 jobs gone, and not from the bottom
We counted every job posted on Upwork in April, May and June: 448,118, down 14.2% on Q1 and the fourth quarterly fall in a row. Inside: reply and view rate by category, budget ranges, what a first reply costs in connects, and the lanes that held up while the market fell.
Fixed-price jobs over $5,000 fell 18.8% quarter on quarter, the steepest fall of any budget band. Jobs under $100 fell just 12.2% and actually gained share of the market.
The quarter in under five minutes
- ▸ 448,118 jobs posted in Q2, down 14.2% on Q1. Fourth straight quarterly decline.
- ▸ The $5K+ band fell hardest at 18.8%. Sub-$100 fell least and gained share.
- ▸ No price compression at the low end. Fewer cheap jobs, priced the same.
- ▸ Expert-level jobs fell 19.1% against 10.9% for entry level.
- ▸ Work needing a live human held flat at 1.6% down and replies at 9.0%.
- ▸ Work that ships as a single file fell 19.9% and replies at 4.2%.
- ▸ Western clients left, South Asian clients did not. France ▼31%, Pakistan ▲6%.
- ▸ A first reply costs $14 in video editing and $50 in web development.
- ▸ A Loom link in the letter replies at 11.3% against 6.0% without one. A GitHub link drops you to 3.3%.
- ▸ March oversold speed. The 15-minute advantage halves once the data matures, and the real cliff is at six hours.
The Upwork funnel in Q2 2026
Upwork view rate and reply rate benchmarks, Q2 against Q1
Every Upwork proposal runs the same gauntlet. A client posts, an agency bids, the client may open it, and only then might they reply.
Both gates tightened. View rate slipped from 18.3% to 17.7% and reply rate from 5.8% to 5.1%.
Neither move is dramatic on its own. What matters is that they fell while the number of jobs to bid on fell 14.2%, so the same proposal volume chased a smaller pool and converted slightly worse.
Is Upwork shrinking? Yes, but not where Upwork said it was
On the Q1 earnings call Upwork gave two reasons for the slowdown. Both are testable against 448,118 job posts, so we tested them.
Four quarters of decline, and it is accelerating
Q3 2025 posted 613K jobs against Q2 2026's 448K, so 27% of the market is gone in nine months. The monthly view shows June was the worst of the six.
Upwork job volume by budget band: every band fell, the top fell hardest
Jobs over $5,000 fell 18.8%. Jobs under $100 fell 12.2%, the shallowest decline on the board.
Read as shares of the fixed-price market, only two bands moved significantly, and they moved in opposite directions: $5K+ lost share (p=0.002) while sub-$100 gained share (p=0.011). The AI-exposed bottom of the market is the part that held up best.
The hourly side refuses to cooperate with a tidy story too. The literal top of the hourly market, $75 an hour and up, fell 16.5% and is statistically indistinguishable from the market.
The band that genuinely broke is $50 to $74 an hour, down 19.5% at p=0.000016. That is the senior-contractor tier, not the elite tier and not the commodity tier.
Test two: is AI making cheap jobs cheaper?
| Sub-$500 fixed price | p25 | Median | p75 | Job count |
|---|---|---|---|---|
| Q1 2026 | $23.22 | $74.90 | $194.78 | 168,593 |
| Q2 2026 | $22.62 | $74.62 | $194.43 | 146,860 |
| Change | -2.6% | -0.4% | -0.2% | -12.9% |
The median sub-$500 job moved by 28 cents. Total dollar volume in the band fell 13.1%, almost exactly in line with the 12.9% fall in job count.
So whatever AI is doing at the low end, it is deleting jobs, not discounting them. The cheap work that still gets posted is priced exactly as it always was.
Fiverr reported the same quarter with revenue down 10% and active buyers down 22% year on year. Two different marketplaces breaking the same way in the same quarter rules out any Upwork-specific explanation as the main driver.
Three things everyone got wrong about the Upwork market in Q2
One of these is a correction to our own March report. We would rather print that than pretend the earlier read still holds.
The freelance market is fine, it is just rotating.
Our own March issue said the market was not shrinking, on the strength of one positive month. Four consecutive quarterly declines and a 27% fall since Q3 2025 say that call was wrong. This is contraction, not rotation.
AI is hollowing out the cheap end of the market.
Sub-$100 jobs fell the least of any band and gained market share, and their price percentiles did not move at all. The band that actually broke is $5,000-plus, down 18.8%, plus the $50 to $74 an hour senior tier, down 19.5%.
Fewer jobs means everyone's reply rate collapses.
Market reply rate fell only 0.7pp to 5.1%, far less than the 14.2% fall in supply. Engineering still replies at 10.3% and telemarketing at 16.1%. Where you bid still matters far more than the weather.
Upwork cover letter data: what actually moves reply rate
We recomputed every structural feature from the March issue on a full quarter of Q2 cover letters, with Q1 alongside. Three of March's findings held, one shrank by half once the data matured, and one turned out to be measuring the wrong thing.
The largest effect anywhere in this report is not speed and not length. A letter containing a Loom or video link replies at 11.30% against 6.04% without one, a 5.26pp gap across 13,522 letters at p<0.0001.
A GitHub link does the reverse and drags you to 3.34%. March reported links as mildly net-negative overall, which was true and also useless, because it averaged the best and the worst thing you can paste into one number.
A 60 second video is worth more than speed, length and word choice put together.
Upwork profile data: what the sending profile changes
Since June 2 we capture a snapshot of the sending profile alongside every automated proposal. This is the first look at it: 53,072 proposals from June 2026, matched to the profile that sent them, across 325 agency accounts and 511 distinct profiles.
Every rate below is computed with the largest sending account excluded, because one agency's house style would otherwise drive the whole table.
Where the eyeballs went
Upwork view rate by category, Q2 2026
Reply rate tells you whether a client wanted to talk. View rate tells you whether they noticed you at all, and the gap between categories is the widest number in this report.
In Engineering, nearly four in ten proposals get opened. In Web Dev, one in seven.
Engineering held at 38.0% while Web Dev slid to 13.8%. That is a scarcity effect, not a quality one, because the same agencies and often the same writers are bidding both.
The mover to watch is Writing, down 6.1pp to 25.6%, the largest view-rate fall of any category. Data Science lost 4.0pp.
View rate across four quarters
Where the conversations started
Upwork reply rate by category, Q2 2026
The ranking barely moved but the levels did. Engineering leads at 10.3%, Customer Service follows at 9.6%, and Web Dev sits at 3.7%.
Writing fell furthest, down 4.2pp from 9.4% to 5.2%. In March it was our second-best category and one quarter later it is barely above the platform average.
Reply rate across four quarters
The saturation paradox
The saturation paradox, one quarter on
March's headline was that Engineering replied 2.7 times more often than Web Dev while taking almost none of the attention. The ratio is now 2.75, up from 2.61 in Q1, so the gap widened slightly.
Web Dev still absorbs 57.4% of all agency bidding and returns 3.7%. Engineering takes 0.6% and returns 10.3%.
Then we tried to break it, and mostly succeeded. The population-level gap is real, but the advice we built on it in March was close to unusable.
So the useful version of this insight lives one level down, at subcategory granularity, where the samples are bigger and the crowding is measurable.
Cost per first reply
Upwork connects cost per first reply, by category
A reply rate on its own is not a budget. Price it in connects and the same finding becomes something you can put in a spreadsheet.
Web Dev costs $51.24 per first conversation. Engineering costs $18.40 and Customer Service $18.20.
Connects prices are similar across categories at $9 to $14 a job, so this ranking is almost purely a reply-rate ranking expressed in money. It is the same fact, in the unit a founder budgets in.
Upwork connects calculator: what your budget actually buys
Put your own monthly connects budget in and pick the lane you bid in. This is the report's central finding applied to your numbers rather than ours.
Upwork subcategory benchmarks: the cheapest lanes to win a reply
This is the table to act on. It adds the two columns an opportunity list normally hides: how many agencies are already bidding, and how much of that bidding belongs to the top five.
Lead Generation and Telemarketing is the best play at scale. It replies at 11.3%, costs $14.85 a conversation, and spreads across 236 different agencies, so no single specialist is propping the number up.
It also has room: roughly six posted jobs for every proposal our customers send, against Web Dev where there are more proposals than jobs. Our knowledge base flagged this same niche two quarters ago, so this is a replication rather than a one-quarter blip.
The oddest row is Blockchain, NFT and Cryptocurrency at 22.4% reply and $8.84 a conversation, the cheapest in the whole market. It sits inside Web, Mobile and Software Dev, the category everyone including us has been telling you to avoid.
It is genuinely tiny, built on 107 proposals from 27 agencies, so treat it as a boutique lane and not a strategy. It does however kill “avoid Web Dev” as a category-level rule, because Web Dev's problem is its generic middle: Web Development costs $50.14 a conversation against $8.84 for this niche inside it.
The fifteen busiest skills, and what happened to each
Web development costs roughly three and a half times more per conversation than video editing, because it pairs a low reply rate with a higher connects price.
This is the same money most agencies are already spending. The difference is whether it buys fifty conversations a month or fifteen.
Budget ranges by category
Upwork fixed-price budget ranges by category
Legal carries the highest median fixed budget at $188, followed by Web Dev at $163 and Engineering at $151. Translation sits at $36.
Narrow bands mean commodity pricing and little room to position. Wide bands mean the same category holds both $50 fixes and $5,000 projects, so the brief you accept matters more than the label.
Client quality patterns
Upwork client quality by category: payment verified, spend and rating
Reply rate is half the picture. Whether the conversation is worth having depends on who is on the other end of it.
The pool got thinner at the top and greener at the entrance. Payment verification fell 2.2pp while first-time clients rose 2.2pp.
Median client lifetime spend rose to $5,397, but read that carefully: when low-spend clients leave, the median rises without anybody spending more.
Hourly versus fixed price
Upwork hourly against fixed-price mix by category
Accounting is 73% hourly and Customer Service 72%, both built for retainers. Design and Writing sit near an even split, which rewards packaged offers.
The mix barely moved quarter on quarter, which is useful: contract structure is a property of the category, not of the market cycle.
Volume sweep by category
Upwork job volume by category across four quarters
Not one of the twelve categories grew. The spread runs from Engineering and Legal at 3.8% down to IT & Networking at 22.1% down.
Web Dev lost the most in absolute terms, shedding roughly 24,500 postings, which is a third of the entire market's loss from one category.
What survived needs a person, not a file
Sort every skill by quarterly change and a pattern falls out that no category chart shows. The winners need a voice, a body, a licence, or a physical object.
Four kinds of work, and how each held up
Group the whole platform by what the work physically requires and the pattern is stark. Work needing a live human held volume almost flat at 1.6% down while the market fell 14.2%.
Pure digital output, meaning anything that ships as a file, fell 19.9%. That is a 18pp spread between the two ends.
The reply rates rhyme. Live-presence work replies at 9.02% and regulated work at 8.01%, against 4.16% for pure digital and a 5.12% market average.
Those reply figures are computed with each group's largest sending account removed, so they are not one specialist agency's house numbers.
The cheapest lanes to buy a conversation in
Videography is the standout, at $8.95 a conversation. It grew 5.8% organically rather than by retagging, and its reply rate rises from 11.64% to 15.69% when the largest agency is excluded, which is the opposite of the concentration problem that killed the other lanes.
It is also barely contested: 44 distinct sending accounts against 7,299 Q2 postings. Note it is a different skill from video editing, which carries eleven times the volume at 81,292 postings, converts worse, and has 27.5% of its bidding concentrated in the top five accounts.
Skills that collapsed
Presentation design fell 38.9%, infographics 37.0%, wireframes and mockups 36.8%, brand guidelines 33.1%, data visualisation 34.1%.
Every one of those ships as a single file, and every one carries a reply rate under 5.3%. The declines and the weak conversion point the same way.
Hourly rate ranges
Upwork hourly rate ranges by category
The ceilings barely moved, which is the interesting part. Clients cut the number of jobs they posted without cutting the rate they offered.
Where the clients went
Upwork job volume by client country, Q2 2026
The contraction has an accent. France fell 31.3%, the Netherlands 25.6%, Canada 18.9%, the UK 18.1%, Germany 17.7% and Spain 13.6%.
Meanwhile India fell only 4.1%, Nigeria 3.6%, and Pakistan grew 6.3%. The clients who stopped hiring are overwhelmingly in high-wage economies.
That fits Upwork's macro explanation far better than its AI one. Tariffs, energy prices and interest rates land on a European mid-market business in a way they do not land on a Karachi startup.
The seniority squeeze
Expert-level work fell nearly twice as fast as entry level
Expert-level postings fell 19.1% while entry level fell 10.9%. Expert share of the market slipped from 26.7% to 25.2%.
This is the same shape as the budget bands, seen from a different angle, and it is the opposite of the story where AI wipes out junior work first.
Upwork platform changes in Q2 2026: a dated timeline
Q2 was one of the busiest product quarters Upwork has had, and two of these changes altered how your proposals are seen rather than how many jobs exist.
- Mar 31Official
Private feedback stopped counting toward Job Success Score ↗
Clients can still leave a private note, but it no longer moves your JSS. The score is now driven by the public star rating alone.
Who it hits: Every freelancer and agency. An invisible input you could never see or contest is gone.
- Apr 9Official
Upwork launched inside ChatGPT ↗
Clients can describe a project, find talent and draft a job post without visiting Upwork, then hand off to the platform to scope and contract it.
Who it hits: A new demand channel that bypasses the normal posting flow entirely.
- May 5Official
Spring 2026 release moved Uma AI shortlisting to the free plan ↗
AI shortlisting used to be a Business Plus feature and is now on Basic. Upwork says clients who use its shortlist make 30% more hires.
Who it hits: Everyone. Far more clients now see an AI-ranked shortlist before they see your proposal.
- May 28Reported
Specialized Profiles were deleted platform-wide
Portfolio items, earnings and work history merged into the main profile automatically. Title, overview and skill list did not transfer and had to be rebuilt by hand.
Who it hits: Anyone who ran niche profiles. One general profile, capped at 20 skills.
- Jun 17Official
Upwork connector launched for Claude ↗
The same pattern as the April ChatGPT integration: describe a project, find talent and create a post from inside the assistant.
Who it hits: Clients, as another discovery surface outside the marketplace itself.
Levit spent the quarter documenting the matching layer rather than the size of the market. Her framing is that the rules changed and freelancers should adapt, which is a real disagreement with this report worth naming rather than glossing over.
Both readings can hold at once. Her audience feels a visibility problem, and visibility is what the specialized-profile deletion and the AI shortlist both changed, while the job count is a separate measurement that fell for its own reasons.
Go deeper on the lanes you choose to fight in
Upwork proposal response rate: how to hit 15%
The eight decisions quietly costing you replies on the funnel.
Read article →Upwork benchmarks for agencies
Reply, shortlist and win rates by budget tier, and what to plan against.
Read article →Upwork revenue growth in 2026
Where the platform is still growing and how to plug an agency into it.
Read article →Three plays for your Upwork pipeline in Q3
Price every lane by cost per first reply
Take last quarter's connects spend per lane and divide by the chats it opened. Rank the lanes. Anything over $40 a conversation has to justify itself with a much better close rate or contract size, and web development is running near $50 while video editing runs near $14. Cut the worst lane and move that budget, not your total budget.
Add one offer that cannot be delivered as a file
Work needing a live human held its volume flat while the market fell 14.2%, and replies at 9.0% against 4.2% for file-shaped work. Videography is the cheapest way in at $8.95 a conversation, with 44 agencies bidding it against 7,299 postings. Telemarketing is the phone-based equivalent at a 16.1% reply rate. Pick one, staff it with one person who can actually show up, and give it four weeks.
Shrink your first ask and follow the demand geography
Expert-tagged jobs fell 19.1% and $5K-plus jobs fell 18.8%, while sub-$100 work gained share and did not get cheaper. Put a paid discovery or audit under $500 at the front of your senior offer. Then reweight the scanner toward India, Pakistan, Nigeria and the Philippines, which held while France, the Netherlands and Canada fell 19% to 31%.
Methodology
Job volume. 13M+ Upwork job postings collected by public crawl since 2023, deduplicated and bucketed by the calendar quarter a job was posted in. Q2 2026 covers April 1 to June 30 and contains 448,118 postings. Same sources and definitions as the March 2026 issue.
View rate and reply rate. Computed from the GigRadar customer-agency proposal pool using Upwork's own definitions: viewed means the proposal was opened, replied means a chat thread opened. Outbound proposals only, invitations excluded.
What changed since March. Three fixes, all of which move published numbers. Proposals whose job carries no category are now excluded from every rate, because that group behaves as a distinct population and was inflating the topline by roughly 1.2pp. Average client rating is now computed only across clients who actually carry a rating, which moves it from about 2.4 to 4.83. Skill trends are now computed on Upwork's stable ontology field with retag families merged, because the legacy skill field lost 15% of its tags per job during the quarter.
Significance. Every quarter-on-quarter claim called a finding here clears a two-proportion z-test at p<0.05. Buckets below the stated minimum sample are reported as insufficient rather than quietly dropped.
Recency bias. Reply rate for the most recent month keeps climbing for 30 to 60 days as late chats arrive, so June 2026 reads low here. March 2026 published at 5.3% and now reads 5.82%, which is that lag in action.
What we did not use. Upwork's own Q2 2026 results were not published when this went out, so no Q2 financials appear anywhere in this report. Where we could not establish a cause we say so instead of guessing.
Vadym Ovcharenko
Founder of GigRadar. Spent 8 years building Upwork agencies before turning the bidder pipeline into a product. Writes these market reports from the same data 2,000+ agencies use to scan Upwork.
A smaller market punishes a bad lane faster.
GigRadar scores every Upwork job by the reply rate of its lane, not by keyword fit. Book a demo and we will price your current mix by cost per first reply against this report.