I stopped asking “is Python still worth learning” about a year ago. The question I get from readers now is sharper, and harder: if I learn Python in 2026, what job will it actually get me, and who will I be competing with for it. That question has a real answer in the survey data, and it is not the headline number you see on most blog posts about this.
What follows is five specific numbers from five specific reports. I read each one against the others so you do not have to. The pattern across them is more interesting than any single stat, and the pattern is what I would actually bet on if I were making a hiring or learning decision today.
The professional share is the number that matters
Stack Overflow polled 31,771 developers in 2025. A little under 58 percent said they had done substantial Python work in the prior year. That is the headline number, and the number most articles stop at. The detail is in the split, and the split is the part that changes what you do with the headline.
Among working developers, the share drops to around 55 percent. Among people still learning to code, it jumps to nearly 72 percent. That gap of about 17 percentage points is the labor market signal. Roughly half of the developers you will sit next to already write Python. Almost three quarters of the people entering the field are picking Python as their first language.
The implication depends on which side of that gap you are on. If you are hiring, the professional share is the metric to plan against. If you are learning, the 72 percent learner number is the one to know, because it tells you who you will be competing against in eighteen months.
- 58 percent of all respondents do substantial Python work, but the headline overstates who is doing it professionally.
- 55 percent of working developers write Python. That is the ceiling on your local candidate pool.
- 72 percent of new learners picked Python. That is the floor on your incoming candidate pool.
- The 17 percentage point gap is the labor market signal. New entrants skew toward Python. Veterans do not.
- Treat the headline as a marketing number and the split as the operational one.
Python 2 is still a real surface area
W3Techs scanned every website with a detectable server-side language on 14 September 2026. Just over one percent of all sites run Python. Of those Python sites, almost 88 percent are on Python 3. That leaves 12 percent still on Python 2.
Python 2 lost official support in 2020. Six years on, more than one in nine public-facing Python servers is running an interpreter that will not get a security patch. That is not a curiosity. That is an attack surface, and the same sites are typically the ones that have not been touched in years for budget or organizational reasons.
If you run a security scan against a server you do not own, and you see Python 2, treat it as a finding worth reporting. If you own a server that runs Python 2, treat the migration as overdue regardless of how stable the service appears. The 88 percent number is good news for the language overall. The 12 percent number is the work that did not get done.
The AI repository count is the directional signal that is hardest to fake
GitHub’s Octoverse 2025 report counted more than half a million repositories tagged as AI projects. That is up more than 50 percent year over year. Python anchors the category. Tags are noisy, and “AI” has been a marketing word since 2023, so the exact number is a ceiling rather than a floor. The directional signal is real though.
PyPI shipped close to 700 million downloads of pandas in the 30 days leading to 14 September 2026. That is roughly two dozen downloads per second of one library. Most of those downloads are not for general scripting. They are for data work, which is mostly AI work now, which is mostly Python. NumPy, scikit-learn, PyTorch, and TensorFlow all assume the caller is writing Python. The strength here is not the language itself, it is the gravitational pull of an installed base that has spent a decade accumulating tutorials, Stack Overflow answers, and library authors.
If you are choosing a language for AI work specifically, the question is not whether Python is dominant. The question is which library set you want to live inside. The answer for most people is still pandas plus one of the deep learning frameworks, and that answer has not changed in five years.
Salary bands are wider than the language choice
The US salary data for Python developers in 2026 clusters in roughly the same shape as last year. Mid-level roles report totals around the same six-figure band they did in 2025. Senior roles with platform or AI specialization sit meaningfully higher. The bands move with location, specialization, and years of experience far more than with which language the person writes in.
Most of the gap between two senior engineers at two different companies is about the company and the city, not the syntax. Recruiters who quote you a number based on “Python demand” rather than your specific role and city are giving you the wrong frame. Self-reported survey data also carries an upward bias that compounds the noise.
Anchor on the median for your role and city, ask the recruiter for the range rather than the top-of-band figure, and discount any quote that depends heavily on the language name. If two candidates with the same role and experience ask for very different numbers, the difference is almost always negotiation skill, not language skill.
- Salary bands cluster tightly across languages. Python and JavaScript are within five percent of each other in 2026.
- Specialization in AI, data engineering, or platform work adds more than language choice does.
- Remote roles shifted the bands. You can now live in a low-cost city and earn a tier-one salary.
- The Python-versus-other-language gap is small. The senior-versus-junior gap is not.
- Treat any recruiter quote above $200,000 as suspect until you see the median for your role and city.
Five other numbers I checked before writing this
A few stats I considered and dropped. The TIOBE index put Python near the top in September 2026, but the year-over-year change shows the whole top tier cooling rather than Python alone. The cooling is not a Python story, it is a top-tier-popularity story, and TIOBE measures search engine attention, not deployed code. Worth knowing, not worth acting on.
PyPI’s overall download count keeps climbing, but it is dominated by transitive dependencies, not human-installed packages, so the number overstates developer activity. The RedMonk language rankings put JavaScript first and Python second, with a near tie for the rest of the top ten. Reddit and Hacker News both trend toward Python for new project discussions, but that is a sentiment number, not an adoption number.
The number I would actually plan around if I were making a learning or hiring decision is the 55 percent professional share. That is the real installed base among people who are paid to write code. Everything else is a multiplier on top of that.
Trade-offs
None of the five numbers is the only number you should care about, and all of them have known biases. The Stack Overflow survey is self-selected, so heavy users overrepresent. The W3Techs web fingerprint misses everything behind a CDN (a content delivery network, the global cache layer that sits in front of most modern sites), so the one percent Python server-side share is a floor, not a ceiling. The GitHub Octoverse tag count is a ceiling too, because not every AI project tags itself. PyPI download counts are dominated by automation. Salary numbers from any source carry 10 to 20 percent upward bias.
The trade-off in interpretation is real. Any single number overstates or understates the truth. The honest read is to look at all five at once, treat the pattern as the signal, and ignore any single one when it disagrees with the others. That costs you precision. It buys you a much better chance of not being wrong about the macro picture.
If you are picking a language to invest the next twelve months in, the answer is still Python for most people, and that answer is supported by all five numbers. If you are picking between Python and a niche language for a specific AI use case, the answer might be different, and you should ignore the popularity numbers entirely. The five numbers are for the generalist case.
What I would tell past me
If I could send three messages back to the version of me that started learning Python in 2014, this is what I would say.
- Pick one library and learn it well before chasing frameworks. pandas is still the right first library for almost everyone. Everything else will outlive it but not replace it.
- The headline number is rarely the operational one. Popularity indices measure attention, not jobs. Survey shares measure self-report, not workforce composition. Read the splits, not the totals.
- Plan around who is already doing the work, not who is learning to do it. The 55 percent professional share is the ceiling on your candidate pool. The 72 percent learner share is the floor on your incoming competition.
The numbers tell you who actually writes Python in 2026. They do not tell you what to do about it, but they tell you who your neighbors and your competition are, and that is the part of the picture worth betting on.
Source: https://fosspost.org/python-statistics/