Hunter Alpha → Xiaomi MiMo-V2.5
41 daysFree for 11 days, then disclosed and repriced. The listing under Xiaomi's own name appeared 30 days after that.
$0.14 in / $0.28 out per million tokens , $0.0028 cached · 1,050,000 token window
Report · 2026-09-22
OpenRouter's anonymous releases are a pattern, not an accident: a frontier-class model appears with no maker and no price, gets claimed and renamed weeks later, and the codename is deleted. This page is the arithmetic over the three we can date — how long each stayed anonymous and what the reveal did to the price.
The codename is not a discount tier and never was. Every model in the line was free while anonymous and paid afterwards, and the price afterwards varies by a factor of 17.9 between the cheapest and the most expensive — so the useful planning assumption is not "it will be cheap", it is "it will be priced by someone else, soon, without notice".
| Codename | Listed | Maker's listing | Days | Turned out to be | Now, per M in |
|---|---|---|---|---|---|
| Hunter Alpha | 2026-03-12 | 2026-04-22 | 41 | Xiaomi MiMo-V2.5 | $0.14 |
| OX Alpha | 2026-08-20 | 2026-08-26 | 6 | Z.ai GLM 5.3 Flash | $0.15 |
| Union Alpha | 2026-09-16 | 2026-09-17 | 1 | Unbiased Pareto | $2.5 |
OpenRouter catalogue reads, dated in the register Verified 2026-09-22 Listed = the day the codename appeared; maker's listing = the catalogue's own created date for the model id that replaced it Two of these dates are different events, and Hunter Alpha is why the table shows both: the identity was disclosed on 2026-03-23, but the listing under Xiaomi's own name appeared on 2026-04-22 — 30 days later. A report that published one "reveal date" per codename would have hidden that.
Two gaps so far: 161 days between Hunter Alpha and OX Alpha, then 27 days between OX Alpha and Union Alpha. Two intervals is not a cadence, and we are not going to fit a curve to it — but the direction is the thing to watch, because a reader deciding whether to build tooling for the next codename is really asking whether there will be a next codename soon.
Free for 11 days, then disclosed and repriced. The listing under Xiaomi's own name appeared 30 days after that.
$0.14 in / $0.28 out per million tokens , $0.0028 cached · 1,050,000 token window
Free while anonymous, repriced at the reveal, then repriced again on 2026-09-22: input up 67%, output up 67%, cached input up 178%.
$0.15 in / $0.5 out per million tokens , $0.05 cached · 1,310,720 token window
Free for two days, then delisted and repriced at 18x the input price of the cheapest model in the line.
$2.5 in / $7.5 out per million tokens , $0.25 cached · 262,144 token window
Free to quote, chart and reprint under CC BY 4.0 with a link to this page. The same rows are at /alpha-line-report.json, and the register they come from is /stealth-models.
How to cite this: Hunter Alpha Hub, “The Alpha Line Report”, hunteralphahub.com/alpha-line-report, read 2026-09-22.
openrouter.ai/api/v1/models, and the whole snapshot is re-checked against that
endpoint by npm run sync-models, which fails on any mismatch rather than warning.
The 2026-09-22 run is why two prices changed on this site the same day.
We can document 3: Hunter Alpha, OX Alpha, Union Alpha. That is a floor, not a total — a codename we cannot tie to a public listing before and after is left out rather than guessed at, and the old codename disappears from the catalogue at the reveal, which makes later verification impossible.
Across the three we can date, the gap from the codename appearing to the maker's own model id appearing was 41, 6, 1 days — a median of 6. The range is the useful part: the fastest was 1 day and the slowest 41.
No. All 3 were free while anonymous and none is free now (0 of 3). The free window is a trial that ends at or before the reveal, and it ends without notice — that is the operational fact to plan around, not the price.
Hunter Alpha was Xiaomi MiMo-V2.5; OX Alpha was Z.ai GLM 5.3 Flash; Union Alpha was Unbiased Pareto. All three are documented from the OpenRouter catalogue, which is where the codenames appeared in the first place.
Between $0.14 and $2.5 per million input tokens — a spread of 17.9x. The codename tells you nothing about the price you will pay afterwards, which is the strongest argument for keeping a fallback configured while you evaluate one.
/stealth-models is the register: one row per codename with the fields a reader re-checks in the catalogue. This page is the arithmetic over those rows — window lengths, cadence, and what the reveals did to the prices — plus the method and the licence. Both are dated; neither is generated from the other.
The line has produced 3 codenames since March 2026 and the gaps are closing. Leave an email and we will send one note when the next anonymous model shows up in the catalogue, with the fields read from the listing.