For the twelve months ending August 2023, American overdose deaths ran to 112,418 on CDC's delay-adjusted basis.
For the twelve months ending February 2026, the figure is 68,641 [1].
The difference is 43,777 people a year, a fall of just under 39 percent.
What that is per day
At the peak this country was losing about 308 people a day. It is now losing about 188 [1].
Year on year the decline is still running: the twelve months to February 2026 are down 12.05 percent on the delay-adjusted basis against the same period a year before, and 13.20 percent on the raw reported basis [1].
Every single drug class fell
CDC's drug-specific data for the twelve months ending December 2025 shows the mechanism, and it is not subtle [2]:
| Drug | Prior year | Latest | Change |
|---|---|---|---|
| Fentanyl | 47,478 | 36,627 | -22.9% |
| Methamphetamine | 27,851 | 24,442 | -12.2% |
| Cocaine | 23,456 | 20,387 | -13.1% |
| Xylazine | 6,109 | 2,483 | -59.4% |
Fentanyl alone accounts for 10,851 fewer deaths in a single year. Xylazine, the veterinary sedative that turned overdoses into limb-threatening wounds, fell by nearly 60 percent. Heroin fell 28.0 percent, bromazolam 54.8 percent, oxycodone 11.4 percent [2].
One caution on that table: these categories overlap. A death involving both fentanyl and cocaine is counted under each. They cannot be added together, and any total built by summing them is wrong [2].
Where it is working, by name
Delay-adjusted, twelve months to February 2026 against the year before [1]:
Rhode Island 309 to 203, down 34.3 percent. Oregon 1,471 to 1,044, down 29.0. Puerto Rico down 25.8. North Carolina 2,729 to 2,065 - 664 people. Alabama down 23.7, Kentucky 23.5, Illinois 20.5, Ohio 2,981 to 2,382.
Where it is not working
Four places went the other way, and leaving them out would make this a worse piece [1].
New Mexico rose from 816 to 1,037, up 27.1 percent. Colorado 1,622 to 1,852, up 14.2. Arizona 2,695 to 2,914, up 8.1. Minnesota 991 to 1,008, up 1.7.
Three of those four are in the southwest. Whatever is working in Rhode Island and North Carolina is not reaching New Mexico.
How to read these numbers honestly
Two things about this dataset are worth knowing before anyone quotes it.
The most recent months are the least complete. CDC footnotes January and February 2026 as "Underreported due to incomplete data" [1]. The newest and best-looking figures will revise upward. That is why every headline number here is the delay-adjusted series rather than the raw count, and why Oregon's raw fall of 40.9 percent is reported above as 29.0 percent instead.
The peak also depends on which basis you use: June 2023 on reported counts, August 2023 on adjusted, at 111,466 and 112,418 respectively [1]. They are close enough that the choice barely matters and different enough that mixing them is sloppy.
A figure of 69,147 for the twelve months ending January 2026 has been circulating. It does not appear anywhere in CDC's current data, which gives 68,669 reported and 69,402 adjusted for that period [1]. The most likely explanation is that an earlier release was revised. The live dataset is the thing to cite.
None of which changes the shape. Forty-three thousand seven hundred and seventy-seven people a year, still here.
