Does Duolingo teach Telugu? I checked its own API

Duolingo's own course route hands back a description for High Valyrian ending "Learn High Valyrian with bite-size lessons based on science." Ask the same route for Navajo and you get that sentence with Navajo in it. Ask it for Telugu and there is no description tag on the page at all. I fetched all three on 3 September 2026. An English speaker can study a language invented for a television series and cannot study Telugu. That is a fact about the catalog, and I have already been through the catalog: nine apps sorted by what you can say, one at a time. The other half of the question is what a Telugu course would do once it arrived, and that half does not depend on Duolingo shipping one.

What Duolingo's own server says when you ask for Telugu

No. There is no Telugu course for English speakers as of 3 September 2026, and two of Duolingo's own endpoints say so independently of each other.

The first is the JSON at https://www.duolingo.com/api/1/version_info, which Duolingo serves publicly. Its supported_directions key maps each interface language to the languages you can learn from it. Under en there are 39 entries: 39 languages available to an English speaker. Telugu's code, te, is not one of them. Hindi's code, hi, is. So are hv and nv, which are High Valyrian and Navajo.

The second is the course route itself. Duolingo's server emits a meta name="description" tag for course pairs that exist and omits the tag entirely for pairs that do not, which makes it a usable test only if you first prove the test discriminates. It does. On 3 September 2026 the Hindi, Navajo, High Valyrian and Spanish course pages all returned that description, worded identically apart from the language name. The Telugu page returned none, and neither did Tamil, Malayalam, Kannada or Bengali. Every one of those nine pages returned HTTP 200, so the signal is the missing tag rather than a missing page.

One more control, because a static shell would fake this result. I requested /course/es/en/Learn-Klingon, a nonsense slug on a pair that really exists, and the server rewrote its own og:url to Learn-Spanish. The route resolves on Duolingo's side. It is not echoing whatever I typed into it.

You can prove me wrong later, and it takes about ten seconds. Open https://www.duolingo.com/course/te/en/Learn-Telugu and view the source. The day that page carries "Learn Telugu with bite-size lessons based on science," a course has launched and this section is out of date.

Duolingo does teach politeness, and that part of the complaint is wrong

Concede this before anything else: Duolingo does treat register as teachable content, and the first version of this post said it could not. Three of Duolingo's own pages falsified that.

On 26 May 2026 its blog published "What Is 'Vosotros' and When Do You Use It?", an article about which second-person form a Spanish speaker picks and who it is for. That is the same category of decision as meeru versus nuvvu. The same blog carries "3 Levels of Politeness in Japanese Formal Language" from 31 May 2023, which opens by saying Japanese has three main formalized levels of formality. Two languages, two articles. Register sits inside the teaching in both.

The other half of my complaint was that the format assumes you can read the target language. Not supported either. Duolingo's own Duolingo 101 post names "Tap what you hear" and "Speak the correct response" among its exercise types, and says that right from the start you complete lessons on speaking, writing, reading and listening. Reading is one mode of four, not the door you have to come through.

The word an exercise cannot see is the person you are talking to

nuvvu and meeru are not two words to learn. They are one decision, the decision belongs to whoever is standing in front of you, and an exercise scored against a single correct string has already made it on your behalf before you finish reading the prompt. A course that introduces the respectful form in one lesson and the casual form in another has taught the wrong unit, however good the exercises are.

Take the strongest reply to that, which the section above already handed me. Put the listener in the prompt. Label an exercise "you are speaking to your manager" and a single-answer drill grades it perfectly well, and Duolingo has published on precisely this choice in Spanish as recently as May this year. Context labels are ordinary. Grading against them is ordinary.

Here is where I think that stops being enough. A labeled prompt tests whether you can apply a rule after being told which rule applies. The thing that goes wrong in a room is not that you did not know the rule. It is that you did not notice the room change: the person you were talking to turns away, their colleague answers instead, and nobody printed a label above her head. Somewhere between the article that explains register and the drill that grades one string, that noticing has nowhere to live.

And the limit on my own claim: I cannot show you what a Duolingo Telugu exercise would do with nuvvu, because there is no Duolingo Telugu course to open. What I can show you is a syllabus built to force the choice instead of labeling it, because I wrote that one.

Six of the fourteen chapters put two registers in one room

Six. In six of the fourteen chapters the learner shares a scene with two people who take different forms of "you," and the first of those lands on day 3. That comes out of chapters[].cast[].varunAddressesAs in the corpus snapshot: count any chapter whose cast holds both a meeru member and a nuvvu member.

Basha drives the auto, Raju runs the tea stall, Venu serves the tiffin: service, and meeru in every case. Rahul has been the learner's closest friend since college, so he gets nuvvu, and using the respectful form on him would read as distance rather than manners. Day 6 is the row worth staring at, because Divya and Joshua are both colleagues on the same team at the same desks, and they take different forms. The variable is closeness, not job title. I made that call in the character file before a line of dialogue existed, and the field sits on the person rather than on the word: varunAddressesThemAs. Six of the nine named characters carry meeru and three carry nuvvu.

It runs back the other way too, and the snapshot records the places it does not know. Divya's 17 lines to the learner carry 14 respectful markers and zero casual ones. Joshua's four lines carry no marker of either kind, so the file records him as unattested, which means no evidence rather than neutral. The corpus gloss for meeru is "you (respectful/plural)", and you can check that against the dictionary entry behind the respectful form instead of taking my word for it.

The sequencing is checkable too. meeku and neeku, the two Telugu words for "to you," both arrive in chapter 4. meeru is taught on day 6 and nuvvu on day 7. The two-register room opens on day 3, three days before either pronoun has a lesson of its own.

Then somebody has to author it and mark it

Grant the objection completely. You can put the listener in the prompt. Somebody still has to write that label for every exercise, decide which form it demands, and mark what the learner typed against it. That is correction. I have priced what an hour of correction actually costs against a month of material, and the two are nowhere near each other. The listener gets authored one prompt at a time or not at all.

Where my own pattern list leans, and it leans away from you

Nineteen sentence patterns carry the whole syllabus, and three of them carry a register marker. All three are the respectful one. Mee daggara ___ unda? opens on the respectful possessive mee; ___ ivvandi. ends on the respectful imperative; ___ ki veltara? ends on the respectful second-person form. Those are the strings as they sit in patterns[].template, blanks and all, and I am leaving the blanks empty here because the blank is the whole point of a pattern. The other sixteen carry no register marker at all. Naaku ___ kaavaali. is a first-person frame and is correct said to anybody.

Zero of the nineteen carry a casual marker. Not one.

Put the taught-side counts next to that. nuvvu is taught in two of the fourteen chapters, s7 and s13, and spoken in the same two. meeru is taught in two, s6 and s10, and spoken in four. Across all 122 phrases the register tags run 48 respectful, 43 neutral, 31 casual.

So if the Telugu you actually need is with your partner, her cousins, and the friends you drink chai with, the pattern inventory in my app leans away from you. The casual half lives in the scenes; it is thinner in the frames you are meant to pick up and reuse. That is a real gap in something I authored by hand, and it costs exactly the person who came looking because they wanted to talk to somebody they love rather than order something from somebody. Weigh that before you start, not on day 9.

What to do while there is no course

Stop waiting. I have no source saying Duolingo will add Telugu and none saying it will not, and anybody who tells you either way is guessing, me included. I can hand you the check instead: open https://www.duolingo.com/course/te/en/Learn-Telugu, view the source, search for a description tag. Present means a course shipped. Absent means nothing has moved since 3 September 2026.

Two things that absence does not mean. It is not a verdict on Telugu, and it is not a finding that Duolingo is bad at what it does. It has a Hindi course. Its blog publishes register teaching in two languages, and its own description of the product names audio and speech exercise types. If Hindi is what you need, go install it today. Coverage is the problem here, not method.

A Telugu course arriving would not settle the argument in the middle of this post either. It would answer the catalog question and leave the format question sitting exactly where it is, because a drill graded against one right string still cannot ask who is standing there. Pick something that makes you choose. Then go be wrong in front of a person, which is the only place the choice ever gets made.