I didn’t believe it at first, but after testing can confirm it’s fully accurate on long context and also very detailed (at least for all the tests I ran)

→ View original post on X — @rowancheung