New audio model snapshots are coming with lower hallucination rates, fewer errors, better instruction following and are stronger at more languages. Testing time

→ View original post on X — @testingcatalog