Comment by ACCount37

5 days ago

Most benchmarks notoriously undercount multi-turn instruction following too. And it's where open source models (and Gemini) lose big time.

0 comments