gpt-oss-120B vs o4-mini: real-world performance on Together AI platform

AI Impact Summary

OpenAI released open-source gpt-oss models (20B and 120B) and pitted them against the proprietary o4-mini on the Together AI platform. In real-world tests, gpt-oss-120B demonstrated solid practical capabilities (e.g., functional snake game generation, strong instruction following) and competitive reasoning, though results varied by task (SVG generation had physics issues). This suggests open-source releases can deliver enterprise-ready performance for common workflows, highlighting potential benefits for on-premises or private-cloud deployments and reduced vendor lock-in; technical teams should assess licensing, deployment options, and integration with existing workflows when considering adoption.