SemiAnalysis posted on X that GPTOSS 2T is a proxy model. The made-up configuration scales the GPTOSS 120B architecture with invented layer counts, hidden dimensions, and expert counts. SemiAnalysis said hardware teams adopt it as a stand-in for closed frontier models such as GPT-5 or Gemini because they cannot access real weights for benchmarking. SemiAnalysis stated this remains standard practice for hardware architecture and bring-up teams.