Full Summary
This Wednesday morning, a major shift in AI is underway, with multiple sources confirming new hardware and specialized models hitting the market. First, both China Daily and News9live report that Chinese AI firm StepFun has launched what it calls the "world's first agentic smartphone," the STEPX Neo. This device runs on a custom operating system and features a personal AI agent named Amoo, designed to understand user intent and proactively complete complex tasks across apps, like planning an entire trip. Social platform Soul is also entering the hardware space with SoulX, a portable AI smart device for emotional companionship, integrating its own large language model. Meanwhile, PrismML has launched Bonsai 27B, a new AI model that News9live and t.co confirm can run locally on smartphones like the iPhone 17 Pro. This compressed model, taking up just 3.9 gigabytes, supports coding, reasoning, and image understanding, generating around 11 tokens per second. This allows advanced AI to run directly on your device, keeping personal data on the phone. In other news, Crypto Briefing indicates Meta AI's model reportedly scored a perfect 30 out of 30 on the Asian Physics Olympiad's theoretical exam, a significant leap in AI reasoning. Legal Reader highlights Luminance's "Luna Crescent," a new vertical AI model specifically for contract work, achieving 5% higher accuracy and generating answers four times faster than general-purpose AI. TechCrunch and WIRED confirm Thinking Machines Lab, co-founded by former OpenAI CTO Mira Murati, has released Inkling, an open-weight AI model designed for customizability. Finally, The Eastern Herald reports that Google DeepMind CEO Demis Hassabis has proposed a FINRA-style self-regulatory body for AI models, staffed by technical experts to review models before release. This means consumers are starting to see AI integrated directly into their devices, offering more personalized and proactive assistance, while businesses gain access to highly specialized AI tools. However, the push for regulation and the complexities of reliable AI development, as seen with Gemini 3.5 Pro's delays, underscore the ongoing challenges in this rapidly evolving field.