Check for AI changes once a week with four public endpoints

Why a check and not a feed habit

The question worth answering weekly is small: did anything in a tool I already use change? Four public endpoints answer most of it, and none of them needs an account.

The point is the fixed list. A short check you repeat beats a timeline you skim, because the comparison you care about is against last week.

The check

GitHub Changelog, https://github.blog/changelog/feed/ - dated product changes for developer tools, including Copilot. The project's own feed reader parsed 10 entries at the last check.

OpenAI News, https://openai.com/news/rss.xml - the vendor's own announcements; the feed carried 1,193 entries at the last check.

arXiv cs.AI, https://rss.arxiv.org/rss/cs.AI - the day's computer-science AI preprints; 634 entries at the last check. Read these as claims to test, never as results.

OpenRouter Models API, https://openrouter.ai/api/v1/models - the model catalogue, 445 entries at the last check, each with pricing in USD per token and a created timestamp. A new model or a price move shows up here first.

What the check will not tell you

Vendor feeds describe features and intent. Not one of them tests the change against your work, and a changelog entry is not a performance result.

Catalogue prices are list prices, and an entry can carry conditional overrides, such as a higher rate above a token threshold. Read the whole pricing object rather than the headline number.

A preprint arriving in a feed says nothing about whether the method works outside its own evaluation.

Keeping the record honest

When you note a change, write the date and the link next to it. When you quote a price, re-check it that day, because the catalogue moves faster than a saved screenshot.

If a feed stops working, treat it as a broken input rather than as a quiet week.