An instrument for AI model releases
AI Model Radar watches the major providers and curated open-source releases, detects when a model is new or changed, and reports the facts — calmly and without hype.
How detection works
On a schedule we poll each provider's official model-list endpoint, normalise every entry into a single canonical schema, and compute a significance hash over the fields that matter (name, status, kind, modalities, context window, max output, pricing, license). When that hash changes, we record a detection with a field-level before/after diff. New models and changed models appear on the feed and can trigger an email alert.
How we pick which models to track
For commercial providers we track every chat model they officially list. For open-source, we follow a set of trusted publishers (DeepSeek, Meta, Qwen, Mistral, Google, Microsoft and more) and surface each lab's most popular flagship plus whatever is trending — there is no size cutoff, so small and large models are equally welcome. We do skip community re-uploads such as quantised, merged and fine-tuned copies. Because discovery favours each lab's headline release, smaller siblings (say an 8B variant) may not appear automatically.
Want a specific model on the radar? Email [email protected] and we'll add it.
How summaries are written
Every model page is built first from deterministic facts. We then gather public context via web search and ask an AI model to write a short, neutral summary. Summaries are AI-generated and always show the sources they drew on. If the AI step is unavailable, the page still renders fully from the facts alone, marked “summary pending”. We never render unstructured AI output as HTML.
Attribution
Specifications come from each provider's own API. Contextual sources are linked on every model page with outbound, no-follow links. Provider names and models are the property of their respective owners; this is an independent registry, not affiliated with any provider.
Privacy
Email alerts use double opt-in: you confirm your address before anything is sent. We store only your email, your provider filter, and the timestamps needed to operate the service. Every email includes a one-click unsubscribe, and you can request deletion at any time. We never share your data.