Tag: #mobile-llm8 results
Bonsai 27B is a new 27-billion parameter large language model designed to run efficiently on mobile devices like the iPhone. It achieves a compact footprint without sacrificing the capabilities typical of its parameter class.
Android is entering its Gemini Intelligence era with deep integration of Google's Gemini AI model into the operating system. This marks a pivotal shift towards on-device AI capabilities for mobile users and developers. A bonus tip highlights using Claude Code as a personal Wall Street analyst.
The Orchard is a free Android app wrapping any LLM in a 13-section reasoning pipeline with local knowledge graph for persistent beliefs, doubts, and goals. It runs entirely on-device with SQLite storage, no servers or data collection. Costs stay flat at scale, resisting prompt injection.
Qwen3.5 2B model runs fully offline on a mid-tier $300 Android phone with tool calling, vision, and reasoning capabilities. No cloud services, accounts, or data transmission required. The app creator demonstrates its unexpectedly strong performance.
Ifanr teases the exposure of an Apple version of ChatGPT, claiming AI-powered Siri will control everything on iPhone. The piece highlights why 50-year-old Apple remains exciting. Full details likely in their WeChat channel.
Qwen released its new 0.8B model, which runs locally on a 7-year-old Samsung S10E at 12 tokens per second using llama.cpp and Termux. After resolving missing C libraries, it handles conversations and complex tasks effectively. This demo highlights powerful tiny LLMs on edge devices.
User praises Gemma 4 E2B as daily driver on Pixel 10 Pro via AI Edge Gallery app, using CPU acceleration at 32K context. Feels 'ridiculously intelligent' like a 7B model, faster than reading speed with function calling. E4B too slow on phone.
New user questions practicality of 2B models like Qwen2.5 and Gemma on smartphones, noting hallucinations in 80% of replies for simple queries. Seeks confirmation if this is expected or user error. Discussion in r/LocalLLaMA.