Sweep
First-party /v1/models endpoints every 10 s, OpenRouter every 20 s, docs pages every 30 s, Hugging Face and lab blogs every minute.
Connecting to the watcher…
We check lab APIs, OpenRouter, Hugging Face and docs pages every few seconds. When a new model shows up, your phone buzzes or rings.
Every model we've spotted, newest first.
Prediction market odds, leaks, and code that lands before a launch.
vLLM, transformers, llama.cpp and SGLang often add a model before it launches. Odds moves of 15+ points in a day show up here too.
Labs update their model lists before they announce. We read those lists.
First-party /v1/models endpoints every 10 s, OpenRouter every 20 s, docs pages every 30 s, Hugging Face and lab blogs every minute.
claude-opus-4-5-20251101, anthropic/claude-opus-4.5 and the docs entry are one model, so you get one alert. Snapshots and quantized re-uploads don't count.
Every channel fires at once: phone push, Pushover alarm, SMS, phone call. Delivery takes under a second.
Takes a minute. Pick your labs, what counts, and how loud. Settings sync to any device.
Sign in with Google or email. No card.
Signed in as
Removes your alerts and history here.
Every plan watches every lab. Paid plans add louder alerts.
$0forever
$4/ month
Most popular
$9/ month
$29/ month
Self-hosting? Every plan is unlocked. SMS, calls and email need your own Twilio/Resend keys and the admin code.
Sources marked "needs key" turn on once the server has a free API key.
| Source | Covers | Every | Last sweep | Status |
|---|
With an API key we catch a new model within 10 seconds (5 on average), and the push arrives under a second later. Lab model lists usually update before the announcement.
We don't know exactly. We show release odds from Polymarket (real money) and Manifold (play money), leaks and "coming soon" reports from the press, Hacker News and lab blogs, and early signals like unreleased models showing up in inference engines. The daily briefing covers all three.
Use Pushover emergency alarms, which repeat until you acknowledge and can use iOS Critical Alerts, or a phone call. ntfy alerts arrive as Time Sensitive and get through any Focus mode you allow them in.
Flagships only: a new model family like gpt-6 or claude-opus-5-5. We skip dated snapshots (-2026-09-01), preview tags, quantized re-uploads (GGUF, FP8, AWQ), embeddings, TTS and moderation models. Everything: any model ID the lab hasn't listed before.
No. The first check of each source only records what's already there. If a source suddenly lists dozens of unseen IDs, we log a notice instead of sending 40 alerts.
No. OpenRouter, Hugging Face, docs pages and lab blogs work without keys. Free keys for OpenAI, Anthropic, Google, Mistral, xAI and others add the fastest source: the lab's own model list.
Your alerts follow your Wozard account.
Tap to pick.
You can change these any time.
Labs publish lots of model IDs. Choose how picky to be.
Only stories that name a model. One alert per story.
Times are in .
One summary a day: what dropped, what's coming, which odds moved.
News waits for the briefing while you sleep.
Turn on one or more. Press Send test to see it arrive.
Every plan watches every lab. Paid plans reach you louder.
We'll alert you when the next model drops.