MiniCPM5-2B on iPhone: blogger tests a local iMessage agent with 6 tools across 8 scenarios
PatrickToulme · x · 2026-09-12
Patrick Toulme argues the compute shortage's solution is local AI on devices people already own — billions of iPhones of untapped compute. Citing OpenBMB's MiniCPM5-2B (2B params, #1 under 4B on Artificial Analysis's index, topping the Agentic Index 20 vs 9, per the Densing Law of capability density doubling every 3.5 months), he tested it as an iMessage agent: 6 tools, 8 real-world scenarios, 3 runs each — handling tasks like "remind me to send Jordan the invoice tomorrow at 9am" with nothing leaving the phone. He believes Fable-quality models will eventually run fully locally on iPhones.
Related event: OpenBMB's MiniCPM5-2B tops sub-4B model rankings and runs offline on iPhone(2 posts)→
More from Infra
- How Different NVIDIA GPUs Change Model Outputs via BF16, FP8, FP4 Precision — Kyrannio · 2026-10-31
- Kubernetes AI Conformance becomes a permanent subproject under SIG Architecture — TerryTangYuan · 2026-10-31
- Google shares first AI-generated response from space — a haiku, per Pichai — LinkedInNews · 2026-10-31
- Debugging llama.cpp: sliding-window context checkpoints silently eat your RAM — jomi-se · 2026-10-31
- Cornell Professor's Startup Makora Wins $1.75M to Make AI Faster and Cheaper to Run — srush_nlp · 2026-10-31
- Transformers.js monthly npm downloads grow 9x in a year, from 1.64M to ~15M — nicodotdev · 2026-10-31