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