How this desk covers model releases

AI Model Pulse is a release desk, not a headline mill. The job is to notice when a lab, cloud, or open-source project actually ships a model or a meaningful serving change — then write what changed, with links you can click.

That sounds obvious. Most "AI news" pages do the opposite. They rewrite a press note, add a generic intro, and publish before anyone has checked the docs. Those pages exist to rank. They do not help a person decide whether to upgrade, wait, or ignore the announcement.

What we read

The first pass is official:

• Lab blogs and system cards (OpenAI, Anthropic, Google, Meta, xAI, Mistral, Cohere)
• Cloud model catalogs (Azure AI, Bedrock, Vertex)
• Hugging Face model and library posts
• GitHub releases for the tools people actually run: Ollama, vLLM, llama.cpp, Transformers

Secondary news is allowed only as a second citation. If the only source is another recap, we do not write.

What has to be true before a post goes up

A post needs a named product or model and a date or version. It needs a primary URL. It has to say what changed in concrete terms: context window, modality, license, region, API surface, weights, or serving stack. If we cannot extract those facts from the source page, we skip the day.

We also cap volume. Two posts in a day is the ceiling. Most days should be one or zero. Empty is a valid outcome. A thin Tuesday post is worse than silence.

How the writing is done

A scheduled Cursor Cloud Agent does the research pass. It is instructed to fail closed, cite sources, and keep a "what we don't know yet" section. That last part matters. Announcements often omit pricing, rate limits, regional availability, or whether the old endpoint still works. Pretending those answers exist is how recap blogs lose trust.

The public site is this Next.js app. Posts live as markdown in the repository. If a piece is wrong, we can unpublish it from the admin screen without waiting for a new deploy of the whole stack.

How to read us

Use this desk as a changelog with context, not as a substitute for the docs. Click the sources. If we got a fact wrong, say so in the comments — on-topic questions get a reply; promo and spam get hidden.

We will not publish rumor roundups, leaked-benchmark theater, or "everything that happened in AI this week" sludge. If nothing shipped, the homepage stays quiet.