Week by week
The chart counts Apps SDK app ids by the week they were created. Every id counted belongs to an app listed in the directory on Oct 13, 2026. The date is when the id was created, not when the app went live: ChatGPT's own date field is the id's creation time, so we label it "created" and never "launched".
The first id dates from the week of Sep 25, 2025. In all, 307 were created in 2025, and the first week with 100 or more was the week of Dec 15, 2025, with 163. From there the pace rose in steps: 2,345 ids in the first half of 2026, then 2,932 in weeks starting July to September alone, even with the last of those weeks still filling in. The typical week grew each quarter: the median settled week held 68 ids in the first quarter of 2026, 110 in the second and 231 in the third.
Counted by the month each week starts in, the busiest month was August 2026, with 1,278 ids, partly because 5 weeks start in it. The steadier measure is the settled weekly median: 244 ids a week over the 8 weeks to Sep 21.
Apps SDK, plugins_ and the rest
Every ChatGPT listing has an id whose prefix marks its family. Apps SDK ids make up 85% of the directory, 899 listings carry plugins_ ids, and the other 103 are split across 5 small families, which we label curated plugin (2 id prefixes), OpenAI connector, connector and templated app. In our listing sample, every Apps SDK listing carries a creation date and no listing from another family does, and the weekly chart covers Apps SDK alone.
The plugins_ family arrived in a different way. Overnight on Oct 8–9, 410 listings in our listing sample appeared for the first time (1,117 store-wide in our Oct 13 analysis), 408 of them with a plugins_ id. Their ids were created as early as Sep 12, 2025 and as late as Oct 3, and the median one was 50 days old when it appeared. That is a migration batch, not a launch wave. The weekly chart counts Apps SDK ids only, so the batch's plugins_ listings are not in it. The report on the Oct 8–9 batch has the details.
Why the last weeks always look empty
The newest bars, the week of Sep 28 (192) and the week of Oct 5 (25), look like a collapse. They aren't one. This is right-censoring: an id created last week enters the count only once its app goes live, and that usually takes weeks.
In our listing sample, 19 apps went live from Oct 7 to Oct 13 and carry a creation date. Their median launch lag was 32.5 days; a quarter took 56.2 days or more, and the longest took 191.1 days. That is a small sample, so read it as an order of magnitude. We treat a week as settled once 3 weeks have passed since it started, so weeks starting on or after Sep 22 are shaded, and they will keep growing as their apps go live. Settled is relative: in our sample, half the apps took longer than 3 weeks, so even settled weeks can keep gaining. The Launch lag report has the full distribution.
Claude grows in bursts
For Claude we count listings by their listing date, not by id creation, and the weekly pattern is different. Over the same 8 settled weeks, its median was 63 new listings a week. Then two weeks stand out: 890 in the week of Sep 28 and 630 in the week of Oct 5. Together that is 1,520 listings, 37% of the 4,141 Claude listings that carry a date. Both weeks also start after Sep 22, so by our own rule they are not settled yet. In our listing sample, 395 of the 398 Claude listings dated the week of Oct 5 are plugins.
Listing dates say when those listings appeared, not why. Compare the stores on settled weeks, and treat a burst as one event rather than a new pace.
How to use the wave
- Read creation weeks as supply. A week's count says how many ids that became live apps were created then. It says nothing about how those apps rank or whether anyone uses them.
- Judge pace on settled weeks. Leave the shaded weeks out of any comparison; they are still filling in.
- Count launches by go-live, not creation. New launches groups ChatGPT apps by the day they went live, and leaves out the Oct 8–9 batch.
- Plan for the lag. The Launch lag page turns a creation date into a likely go-live window.
- Watch the totals settle. Market pulse shows weekly launches for both stores, with the provisional weeks marked.