About
How the service works
DROPCON is a lead score read from Polymarket's release odds, with weights set by hand. Here is the arithmetic, and the rest of the fine print.
How the score adds up
- 45Next Claude Haiku (4.6+) (Anthropic) · 57% within 7 days, interpolated 7 Oct → 15 Oct (80 × 0.567 = 45.3)
- 4Next Fable (5.2+) (Anthropic) · 93% within 30 days, 36% above the best 7-day read (Next Claude Haiku (4.6+), Anthropic) (10 × max(0, 0.931 − 0.567) = 3.6)
- 3Best 7-day odds moved +8 pts in 24 hours (full marks at +30) (10 × clamp(0.082 / 0.3, 0, 1) = 2.7)
- 52= LEAD SCORE out of 100; the rows above add up to it
Markets put a named frontier release at roughly even odds within 7 days.
- 1RELEASE SURGE75+
- 2MARKETS SMELL A DROP55–74
- 3GPU FANS SPINNING35–54
- 4RUMOUR MILL15–34
- 5QUIET ORBITbelow 15
Weights 80 / 10 / 10, set by hand.
Is this a forecast?
No. It's a hand-weighted lead score, not a probability: a fitted forecast was tested and did not beat the base rate. The evidence is on the backtest.
Service history · last 7 days
DROPCON lead score, hourlyNOW · 3 OCT 15:28Z
The line is DROPCON's lead score, 0 to 100, read hourly from Polymarket's release odds; its band sets the level, 5 (quiet) to 1 (release surge).The line is the 0–100 lead score; its band sets the level, 5 quiet to 1 surge. How it adds up · The backtest
lead scoreAa line's launch, first listed on OpenRouter: context, not scoredolder scoring, not comparable
The week as a table
| HOUR (UTC) | SCORE | LEVEL |
|---|---|---|
| NOW · 3 Oct 15:28Z | 52 | 3 |
| 3 Oct 15:00Z | 52 | 3 |
| 3 Oct 14:00Z | 51 | 3 |
| 3 Oct 13:00Z | 51 | 3 |
| 3 Oct 12:00Z | 52 | 3 |
| 3 Oct 11:00Z | 51 | 3 |
| 3 Oct 10:00Z | 51 | 3 |
| 3 Oct 09:00Z | 51 | 3 |
| 3 Oct 08:00Z | 51 | 3 |
| 3 Oct 07:00Z | 48 | 3 |
| 3 Oct 06:00Z | 47 | 3 |
| 3 Oct 05:00Z | 49 | 3 |
| 3 Oct 04:00Z | 49 | 3 |
| 3 Oct 03:00Z | 56 | 2 |
| 3 Oct 02:00Z | 56 | 2 |
| 3 Oct 01:00Z | 54 | 2 (held) |
| 3 Oct 00:00Z | 45 | 3 |
| 2 Oct 23:00Z | 45 | 3 |
| 2 Oct 22:00Z | 45 | 3 |
| 2 Oct 21:00Z | 45 | 3 |
| 2 Oct 20:00Z | 44 | 3 |
| 2 Oct 19:00Z | 44 | 3 |
| 2 Oct 18:00Z | 46 | 3 |
| 2 Oct 17:00Z | 45 | 3 |
| 2 Oct 16:00Z | 45 | 3 |
| 2 Oct 15:00Z | 43 | 3 |
| 2 Oct 14:00Z | 43 | 3 |
| 2 Oct 13:00Z | 43 | 3 |
| 2 Oct 12:00Z | 42 | 3 |
| 2 Oct 11:00Z | 42 | 3 |
| 2 Oct 10:00Z | 42 | 3 |
| 2 Oct 09:00Z | 41 | 3 |
| 2 Oct 08:00Z | 41 | 3 |
| 2 Oct 07:00Z | 46 | 3 |
| 2 Oct 06:00Z | 44 | 3 |
| 2 Oct 05:00Z | 43 | 3 |
| 2 Oct 04:00Z | 42 | 3 |
| 2 Oct 03:00Z | 23 | 4 |
| 2 Oct 02:00Z | 25 | 4 |
| 2 Oct 01:00Z | 23 | 4 |
| 2 Oct 00:00Z | 55 | 2 |
| 1 Oct 23:00Z | 63 | 2 |
| 1 Oct 22:00Z | 63 | 2 |
| 1 Oct 21:00Z | 61 | 2 |
| 1 Oct 20:00Z | 62 | 2 |
| 1 Oct 19:00Z | 62 | 2 |
| 1 Oct 18:00Z | 34 | 3 (held) |
| 1 Oct 17:00Z | 38 | 3 |
| 1 Oct 16:00Z | 49 | 3 |
| 1 Oct 15:00Z | 44 | 3 |
| 1 Oct 14:00Z | 43 | 3 |
| 1 Oct 13:00Z | 43 | 3 |
| 1 Oct 12:00Z | 43 | 3 |
| 1 Oct 11:00Z | 54 | 3 |
| 1 Oct 10:00Z | 54 | 3 |
| 1 Oct 09:00Z | 44 | 3 |
| 1 Oct 08:00Z | 43 | 3 |
| 1 Oct 07:00Z | 42 | 3 |
| 1 Oct 06:00Z | 45 | 3 |
| 1 Oct 05:00Z | 52 | 3 |
| 1 Oct 04:00Z | 53 | 3 |
| 1 Oct 03:00Z | 53 | 3 |
| 1 Oct 02:00Z | 53 | 3 |
| 1 Oct 01:00Z | 43 | 3 |
| 1 Oct 00:00Z | 43 | 3 |
| 30 Sep 23:00Z | 43 | 3 |
| 30 Sep 22:00Z | 43 | 3 |
| 30 Sep 21:00Z | 45 | 3 |
| 30 Sep 20:00Z | 43 | 3 |
| 30 Sep 19:00Z | 43 | 3 |
| 30 Sep 18:00Z | 38 | 3 |
| 30 Sep 17:00Z | 43 | 3 |
| 30 Sep 16:00Z | 37 | 3 |
| 30 Sep 15:00Z | 36 | 3 |
| 30 Sep 14:00Z | 47 | 3 |
| 30 Sep 13:00Z | 63 | 2 |
| 30 Sep 12:00Z | 69 | 2 |
| 30 Sep 11:00Z | 70 | 2 |
| 30 Sep 10:00Z | 70 | 2 |
| 30 Sep 09:00Z | 67 | 2 |
| 30 Sep 08:00Z | 70 | 2 |
| 30 Sep 07:00Z | 63 | 2 |
| 30 Sep 06:00Z | 62 | 2 |
| 30 Sep 05:00Z | 71 | 2 |
| 30 Sep 04:00Z | 70 | 2 |
| 30 Sep 03:00Z | 70 | 2 |
| 30 Sep 02:00Z | 70 | 2 |
| 30 Sep 01:00Z | 69 | 2 |
| 30 Sep 00:00Z | 69 | 2 |
| 29 Sep 23:00Z | 69 | 2 |
| 29 Sep 22:00Z | 69 | 2 |
| 29 Sep 21:00Z | 54 | 3 |
| 29 Sep 20:00Z | 48 | 3 |
| 29 Sep 19:00Z | 65 | 2 |
| 29 Sep 18:00Z | 65 | 2 |
| 29 Sep 17:00Z | 59 | 2 |
| 29 Sep 16:00Z | 59 | 2 |
| 29 Sep 15:00Z | 58 | 2 |
| 29 Sep 14:00Z | 57 | 2 |
| 29 Sep 13:00Z | 57 | 2 |
| 29 Sep 12:00Z | 39 | 3 |
| 29 Sep 11:00Z | 28 | 4 |
| 29 Sep 10:00Z | 28 | 4 |
| 29 Sep 09:00Z | 44 | 3 |
| 29 Sep 08:00Z | 29 | 4 |
| 29 Sep 07:00Z | 29 | 4 |
| 29 Sep 06:00Z | 54 | 3 |
| 29 Sep 05:00Z | 29 | 4 |
| 29 Sep 04:00Z | 29 | 4 |
| 29 Sep 03:00Z | 29 | 4 |
| 29 Sep 02:00Z | 29 | 4 |
| 29 Sep 01:00Z | 28 | 4 |
| 29 Sep 00:00Z | 28 | 4 |
| 28 Sep 23:00Z | 28 | 4 |
| 28 Sep 22:00Z | 28 | 4 |
| 28 Sep 21:00Z | 38 | 3 |
| 28 Sep 20:00Z | 38 | 3 |
| 28 Sep 19:00Z | 43 | 3 |
| 28 Sep 18:00Z | 28 | 4 |
| 28 Sep 17:00Z | 80 | 1 |
| 28 Sep 16:00Z | 77 | 1 |
| 28 Sep 15:00Z | 78 | 1 |
| 28 Sep 14:00Z | 79 | 1 |
| 28 Sep 13:00Z | 78 | 1 |
| 28 Sep 12:00Z | 81 | 1 |
| 28 Sep 11:00Z | 75 | 1 |
| 28 Sep 10:00Z | 74 | 2 |
| 28 Sep 09:00Z | 75 | 1 |
| 28 Sep 08:00Z | 75 | 1 |
| 28 Sep 07:00Z | 75 | 1 |
| 28 Sep 06:00Z | 75 | 1 |
| 28 Sep 05:00Z | 75 | 1 |
| 28 Sep 04:00Z | 76 | 1 |
| 28 Sep 03:00Z | 77 | 1 |
| 28 Sep 02:00Z | 75 | 1 |
| 28 Sep 01:00Z | 78 | 1 |
| 28 Sep 00:00Z | 79 | 1 |
| 27 Sep 23:00Z | 80 | 1 |
| 27 Sep 22:00Z | 82 | 1 |
| 27 Sep 21:00Z | 77 | 1 |
| 27 Sep 20:00Z | 83 | 1 |
| 27 Sep 19:00Z | 79 | 1 |
| 27 Sep 18:00Z | 76 | 1 |
| 27 Sep 17:00Z | 75 | 1 |
| 27 Sep 16:00Z | 75 | 1 |
| 27 Sep 15:00Z | 85 | 1 |
| 27 Sep 14:00Z | 77 | 1 |
| 27 Sep 13:00Z | 73 | 2 |
| 27 Sep 12:00Z | 75 | 1 |
| 27 Sep 11:00Z | 82 | 1 |
| 27 Sep 10:00Z | 84 | 1 |
| 27 Sep 09:00Z | 87 | 1 |
| 27 Sep 08:00Z | 87 | 1 |
| 27 Sep 07:00Z | 77 | 1 |
| 27 Sep 06:00Z | 78 | 1 |
| 27 Sep 05:00Z | 73 | 2 |
| 27 Sep 04:00Z | 71 | 2 |
| 27 Sep 03:00Z | 71 | 2 |
| 27 Sep 02:00Z | 71 | 2 |
| 27 Sep 01:00Z | 71 | 2 |
| 27 Sep 00:00Z | 67 | 2 |
| 26 Sep 23:00Z | 67 | 2 |
| 26 Sep 22:00Z | 58 | 2 |
| 26 Sep 21:00Z | 58 | 2 |
| 26 Sep 20:00Z | 55 | 2 |
| 26 Sep 19:00Z | 62 | 2 |
| 26 Sep 18:00Z | 62 | 2 |
| 26 Sep 17:00Z | 71 | 2 |
| 26 Sep 16:00Z | 71 | 2 |
| 26 Sep 15:00Z | 57 | 2 |
| LISTED (UTC) | LAUNCH | LAB |
|---|---|---|
| 29 Sep 17:28Z | GPT-6.1 Sol / GPT-6.1 Sol Pro | OpenAI |
| 28 Sep 18:04Z | Claude Sonnet 5.5 | Anthropic |
Questions riders ask
What is whenmodel?
A dashboard of free public signals about frontier AI model releases. Its headline level, DROPCON, reads Polymarket release odds only. Model registry listings, lab blogs, SDK changelogs, Hacker News, stealth slots and leaks are shown beside it with their track records and never move it.
How is DROPCON computed?
From Polymarket release odds alone. For each frontier lab it reads the "released by" curves of its named text-model families: 80 × the best 7-day probability, plus 10 × whatever the best 30-day probability adds, plus up to 10 when the 7-day odds jumped in the last 24 hours. 5 is quiet, 1 is a release surge. Reads extrapolated to a rung more than 14 days out are shown but never scored. The weights are hand-set and the level itself has never been tested. In the same replay, the best 7-day read across labs, the input behind 80 of the 100 points, scored worse than the base rate at 7 days (Brier skill −1.48 (95% interval −3.36 to −0.28)). Early warnings and launches that already landed are listed with their track record and never move the level. What was tested, and what was not.
Where does the data come from?
Polymarket (gamma API), OpenRouter (models API), Hugging Face (trending models and daily papers), Hacker News (Algolia search, including a leak search), TestingCatalog, the labs' YouTube channels, the transformers model registry, the OpenAI and DeepMind RSS feeds, the Anthropic newsroom, xAI's release notes and GitHub SDK releases. All free, all public, all fetched at the edge.
What can't it do?
Mostly, see a launch coming. Registry listings, feeds and Hacker News confirm a launch as it happens, and the markets that reached 90% early had mostly followed a public leak or notice. The backtest measures the signals against real launches; it does not score the level. Read the backtest.
Why no X / Twitter feed?
There is no free read API any more, and every mirror worth using has been shut down. We link the accounts that leak first instead. If that changes, the feed slots straight in.
How often does it refresh?
Each source is cached at the Cloudflare edge for 5 to 30 minutes depending on how fast it moves. The open page checks for new data every 5 minutes and offers a one-click reload; it only reloads itself automatically once you've been idle for a minute with nothing expanded.
Is this financial advice?
No. Odds are what traders think, not what labs have decided. Model release dates slip. Do not bet the GPU budget on a DROPCON reading.
A brief history of vague-posting
- MAY 2024
"her"
One word from Sam Altman, posted 17 minutes after the GPT-4o announcement hit Hacker News, not the night before as this page once said. The vague-post became the genre anyway. Ever since, a lowercase tweet with no punctuation has moved more evals than a press release.
Check the timestamps - AUG 2024
STRAWBERRY SEASON
A photo of strawberries, a "Q*" rumour cycle, and weeks of strawberry replies. o1 shipped in September. One photo and one launch make an anecdote, not a lead time.
- 2025
THE ARENA STEALTH ERA
Anonymous checkpoints with codenames turned up on LMArena around launches: "nano-banana", "gemini-test", assorted mystery models. Not measured here: the one arena registry this site could check, Design Arena, purged its codenames in June 2026 and has listed each one only after launch since.
What didn't lead - 2025
THE CHANGELOG LEAK
Model IDs turn up in SDK release notes, config strings and API error messages. Measured against the announcement, SDK releases land with it or after it: 5 of 7 lagged, and the best, Opus 5.5, beat the post by two minutes. GitHub release feeds were supposed to be the new pizza delivery van. They turn up once the party has started.
What didn't lead - 2025–26
MARKETS PRICE THE DROP
Polymarket started listing "released by" markets per lab. Traders now aggregate every rumour into one number. When a lab's odds crossed 50% with a day to spare, the model shipped on time 12 of 23 times.
The backtest - NOW
WHENMODEL
Release markets, OpenRouter listings, Hugging Face trending, Hacker News, lab blogs and SDK changelogs in one place. DROPCON reads the markets; the rest is shown, not scored.
Source health
16/16 sources answered this build. Each panel's pill reads the health of its own sources.
- Up: Polymarket
- Up: OpenRouter
- Up: HF trending
- Up: HF papers
- Up: Hacker News
- Up: OpenAI news
- Up: DeepMind blog
- Up: Anthropic news
- Up: xAI news
- Up: GitHub SDKs
- Up: HN leaks
- Up: TestingCatalog
- Up: HN launch stories
- Up: YouTube broadcasts
- Up: transformers registry
- Up: First-seen ledger
Disclaimer
This dashboard is for informational and entertainment purposes only. Prediction market odds are crowd opinion, not lab roadmaps. Listing dates on model registries can lag or lead official announcements. Model names, labs and trademarks belong to their owners; whenmodel is not affiliated with any of them, nor with the very good site that inspired it, pizzint.watch. Always monitor the changelogs responsibly.
SOURCES Polymarket OpenRouter HF trending HF papers Hacker News OpenAI news DeepMind blog Anthropic news xAI news GitHub SDKs HN launch stories HN leaks TestingCatalog YouTube broadcasts transformers registry