Your AI news feed is repeating vendor benchmarks nobody checked. We built one that labels them.

Agentic Weekly — Thu, 29 Oct 2026 17:56:30 +0000

A new model drops. Within an hour, a hundred posts say it "beats GPT" — on a benchmark the vendor picked, ran, and published about themselves.

Almost nobody says the quiet part: nobody checked.

We got tired of losing an hour per release digging through docs, pricing pages and model cards just to answer one question: what actually changed, and does it matter to me? So we built Agentic Weekly around one rule.

Every entry carries a label

Reported — we read the primary sources (docs, changelog, pricing page, model card) and every fact links straight to where it came from. Vendor benchmark numbers are marked as vendor claims. We did not run it.
Tested — we actually ran it. A small smoke test, not a benchmark, with the scope, date and log published. Failures included.

Most entries are Reported, and the page says so. We'd rather show you a gap than fill it with a confident guess. When we get something wrong, the fix is dated in a public corrections log.

Built for your agents, not just your eyes

This is the part devs tend to like. The site is designed to be read by machines as much as people:

An MCP server at https://theagenticweekly.com/mcp with read-only tools: latest_models, get_model, latest_releases, compare_tools. It's in the official MCP registry as com.theagenticweekly/agentic-weekly, so your agent can ask "what shipped this week?" and get sourced answers.
llms.txt plus a Markdown version of every page, so LLMs don't have to scrape HTML.
Open data: JSON and CSV under CC BY 4.0. Build on it.

Why it's different

Most AI news sites compete on speed and hype. We compete on one thing: telling you exactly how much we know about each claim, and where it came from. Every model page has its release date, price, context window, licence and modalities, each with a source link. Every release brief lists what changed, what the source says, and what we couldn't verify.

No signup to use any of it.

The Monday email

If you want it in your inbox: a free Monday email with the week's few big releases and 5 short use cases for the new models, each with a prompt you can paste (clearly labelled as suggestions we haven't tested).

We want you to try to break it

Is Reported vs Tested actually useful to you, or noise?
What should we test first?
Find something wrong on a page you know well. Drop it in the comments and we'll fix it and log it publicly.

👉 theagenticweekly.com

New model "beats GPT"? On a benchmark the vendor ran on itself. Nobody checked. Agentic Weekly labels every AI release Reported (sourced, not run) or Tested (we ran it, logs public). MCP server + open data. Try to break it: theagenticweekly.com

Agentic Weekly — Thu, 29 Oct 2026 17:53:08 +0000