01wowsuchai
Before you build on a new AI model, read its sources
For developers picking AI models and tools for a product. Each day and each week, wowsuchai lists the new models and tools it found. Each one links to a record of what its sources say it does and how you get it, with the source behind every claim.
Free, no account. A daily brief after each UTC day closes, a weekly brief every Monday.
02Problem
Launch posts don't tell you what you need to check
A new model comes out every week, each one called the best. The benchmark chart doesn't tell you whether you can get the weights or run it on your own servers. The search results are listicles. And the part that decides whether you can use it sits in the small print: a model can have open weights and a non-commercial licence, which may rule out the product you sell.
wowsuchai pulls those claims out of the maker's pages and puts each one next to its source. You still read the source before you rely on it, but you start from one page.
03The briefs
What wowsuchai found, one page a day
- Each daily brief lists up to 12 new models, developer platforms and apps that wowsuchai found that UTC day. Each weekly brief lists up to 25. On a busy day, the rest are left out of the brief and stay in the catalog.
- A brief lists what was found, not what was released that day. A model released earlier, for example on Hugging Face, shows up when it is found.
- Each item gives the maker, what it does, how you can get it (open weights, API, CLI), its sources and an evidence score. Items are ranked by evidence score: how well the sources back the claims, not how good the model is.
- A model writes the short summary at the top, and code checks that every sentence in it points to the items it is about. No person approves it. Nothing in a brief is paid for.
- Corrections to earlier records are listed too.
- Briefs also come by RSS or email.
04Records
Claims next to their sources
Open any item and you get its record. Each claim on it, like what the tool does or where you can get it, links to the page it came from. Most also show the quoted passage. The record says whether that page is the maker's own or an independent one, and when it was last checked.
"This model is released under the TimesFM Non-Commercial License v1.0."
An automatic check marks whether each quote supports its claim, and it can be wrong. The source is often the maker itself. Each record also has an evidence score from 0 to 1: how well its sources back it, higher when the maker's own page does. A record with no source from its maker can't score high, and records at the lowest level stay visible, marked low confidence. When a source changes, the record is corrected and the earlier version is kept.
Records are written from the makers' own announcements, model cards and docs. The sources include labs like OpenAI, Anthropic, Google, Mistral, Qwen, DeepSeek and NVIDIA, plus new models on Hugging Face and new repositories on GitHub. wowsuchai reads 87 sources every 6 hours and publishes without a person checking each record, which is why every claim links to its source.
05What you still do
What wowsuchai doesn't do for you
- It doesn't measure how well a model performs. Test it on your own task.
- It shows licence terms when it finds them, and many records have none. It doesn't tell you what a licence means for your product.
- It rarely says what a hosted tool does with your data. Read the provider's terms.
- Labels, summaries and quotes can be wrong, and not every claim has a quote. Open the source before you rely on one, licences especially.
- It doesn't watch every lab. A release announced only on a site it can't read can be missed.
06More ways in
Search the records, or ask what to use
wowsuchai started collecting on 23 September 2026. Most older tools, like Whisper or GPT-4o, are not in it.
- Search the catalog and filter by capability, open weights, API or on-device.
- Put 2 to 4 records side by side. Each requirement you set shows met, not met or unknown, with its evidence, and unknown is common.
- Describe what you're building and get a suggested setup: the models and tools from the catalog that fit, or an answer that says what evidence is missing. The capability must come from a verified source; where it runs and what it costs come from the record's labels.
- Agents can read everything through a JSON API with no key.
07Price
Price
Reading is free and needs no account: briefs, records, search, comparisons and the API.
On the website, 3 suggested setups a day are free. After that, each answer costs $0.05: sign in and buy 100 for $5, or pay per call on the API. An answer that says the evidence is missing still counts as one. If it turns your request down or fails, you aren't charged.
08Read
Start with the latest brief
The daily and weekly briefs are listed newest first. Open any item to see its sources.