According to Paolo Ardoino, QVAC Workbench is running SmolVLM2 on his phone with approximately 30 tokens per second throughput and about 1.5 seconds time-to-first-token, fully private and on-device, providing concrete mobile inference performance data. Source: Paolo Ardoino on X (Nov 2, 2025). These verified metrics (30 tok/s, ~1.5s TTFT) offer a real-world baseline that traders can use to benchmark claims from AI-crypto projects marketing on-device or edge inference capabilities, especially those emphasizing privacy-first design. Source: Paolo Ardoino on X (Nov 2, 2025). For crypto market context, the disclosure of low-latency, on-device VLM inference may focus attention on AI-related tokens and infrastructure plays tied to edge AI narratives, where comparable mobile performance figures become a due-diligence checkpoint. Source: Paolo Ardoino on X (Nov 2, 2025).