A developer has successfully run a 90M-parameter conversational LLM on a Sony PSP, hardware originally released in 2004. The model operates at approximately 0.5–0.6 tokens per second, making responses take 1–3 minutes, but remains functional for generating poetry and short stories. The project is open-source on GitHub.