New GPT-4 Turbo needle-in-haystack long context benchmark:

→ View original post on X — @gdb