OpenAI released its first open-source language models in over six years — the gpt-oss series (20B and 120B parameter reasoning models). Together AI ran five practical tests comparing gpt-oss-120B against o4-mini: code generation (snake game), SVG creation, instruction following, math reasoning, and web-enhanced research. gpt-oss-120B scored 4.5/5 vs o4-mini's 3/5. The post highlights open-source advantages including Apache 2.0 licensing, fine-tuning freedom, and significantly lower inference costs, while promoting Together AI's platform as the place to run these models.
