AI model news

The Modelglass registries record what is true now — a model's current price, its verified capabilities, its benchmark scores, each entry sourced and dated. Model decisions are rarely only about the present, though. A provider announces a price rise that starts next month. A new frontier model is dated but not yet released. A capability is flagged for removal. None of that belongs in a registry built to describe the current state.

The news feed is that forward layer. It captures the delta — what is changing, and when — in the window before a change is codified into the pricing and capability registries. The scope is deliberately narrow: frontier model announcements, planned changes to models already tracked, and pricing changes that have not taken effect yet. Funding rounds, product marketing, and opinion pieces are not signals we record.

The hard part is the filter. AI coverage is relentless, and most of it is noise for someone trying to make a routing decision. Getting the feed to carry only entries that matter — each tied to a primary announcement and a specific model or provider — took a lot of refinement. It runs on the same daily cadence as the other registries, with the same sourcing discipline.

For anyone wiring model choice into a pipeline, this is the input that lets a system get ahead of a change instead of reacting to it. A price increase two weeks out, or a model being sunset, is a decision worth making before it lands — not after.