ArtificialWatch: Know the minute a new AI model drops.

Hey Product Hunt 👋 ArtificialWatch started as a much smaller site: a tracker for one model. When Anthropic's Fable 5 was suddenly pulled from Claude plans earlier this year, I built FableWatch to follow the saga – and the thing people kept asking for wasn't analysis. It was: "just tell me the second it's back." Then I noticed the same problem everywhere. AI labs ship their biggest releases with zero warning – a model ID quietly appears in an API catalog, and by the time it's on your feed, the thread is 6 hours old. This week alone: Claude Opus 5 was spotted on Vertex AI, Kimi K3's launch page leaked and vanished, and Gemini 3.5 Pro leaks point to Thursday. Nobody should have to doomscroll X to catch this. So ArtificialWatch sweeps the providers' public APIs every 60 seconds. When a watched model appears, we don't alert on a rumor — the pipeline makes the model answer a real request first, then fires. Email and push on the free tier; paid tiers get SMS and an actual phone call (sounds ridiculous until your competitor started building on the new model 6 hours before you — a text gets missed, a call gets answered). A few things I'm proud of that are all free and public: 📡 The radar — 43 models tracked, with live Polymarket odds on when each drops 📊 The Bench Index — every major leaderboard (LMArena, LiveBench, SimpleBench + more) in one normalized table 📉 The Drift Index — daily verified checks answering "is my model quietly getting worse?" 🧾 Our public track record — every call we made, timestamped, misses included That last one is the whole philosophy: an alerts product only works if you trust it, so everything we claim is verifiable on the site. Would love your honest feedback — especially: what model or surface should we watch that we don't yet? I'll be in the comments all day. 🚀

Wispr Flow: Dictation That Works EverywhereStop typing. Start speaking. 4x faster.

Promoted

Maker

📌

16h ago

A phone call alert for a new model drop feels like overkill and could get annoying fast. How about letting users set quiet hours or at least choose the alert channels per provider so the noise stays manageable.

2h ago

swept OpenAI and Anthropic in under a minute and pinged me before the official blog post landed. the phone call alert is overkill but oddly reassuring for a dropped model i actually care about.