The smolmaxxing project achieves localized inference on budget chips by relying on optimized memory management and on-demand flash paging rather than virtual simulations. Developers successfully operated models reaching up to 16 billion parameters across non-rooted streaming boxes and decade-old tablets with only 1 GB of usable RAM. Future updates will broaden support across additional legacy routers, single-board computers, and older mobile phones.
Key points
Custom kernels and weight streaming enable multi-billion-parameter systems to function wi…: Custom kernels and weight streaming enable multi-billion-parameter systems to function within severe hardware limitations.
Direct hardware evaluation validates that older devices with 1 GB of RAM can support loca…: Direct hardware evaluation validates that older devices with 1 GB of RAM can support local text generation.
Developers aim to extend the optimization framework to run on consumer routers and single…: Developers aim to extend the optimization framework to run on consumer routers and single-board computers.
Sources checked by our analysis pipeline: 1
Every device around you is more capable than it looks. TV boxes, old phones, single-board computers and routers all ship with CPUs, RAM and flash that sit idle most of the day. smolmaxxing is about unlocking that capability: finding the biggest language model each device can really run, then pushing it further with custom kernels, smarter memory use and weight streaming. Every claim is measured on the hardware itself.
Device #1 is a Xiaomi TV Box S (3rd Gen): Amlogic S7D, 4× Cortex-A55 @ 2.5 GHz, 1.9 GB of RAM (about 1 GB usable), eMMC flash, Android 14, no root. On it, this repo runs models from 260 K to 16 B parameters:
- Qwen3.5-0.8B: 16.8 tok/s (2.2× the stock 7.5; measured in the evening of 11-05, see §17).
- Qwen3.5-2B: 7.7 tok/s in real chat. The box's speed drifts by ~10 % over a day, so every step is measured back to back: §16 gave +11 %, §17 another +7 %.
- An 8B mixture-of-experts model (Moon-Beam-8B-A1B): 7.2 tok/s in real chat, its experts paged from flash on demand…
The leadership shakeup at Nvidia comes as the chipmaker works toward a pending $12.9 billion takeover of the popular artificial intelligence repository. Clément Delangue's pledge seeks to calm industry fears after losing a key open-weight advocate, especially with the transaction facing antitrust scrutiny over model distribution dominance.
The carrier is updating its ticketing and distribution policies across its proprietary channels, travel agencies, and aggregators to bar unauthorized third-party automated tools. The move addresses growing corporate difficulties in distinguishing between real customers and automated bots while shielding traveler data from leaks. By taking this step, the airline aims to eliminate compliance hazards tied to mandatory passenger safety disclosures.
Modern battlefields require forces to act with speeds that deny adversaries time to respond effectively. Addressing senior commanders in New Delhi, Defence Minister Rajnath Singh urged India's military leadership to incorporate cyber, space, and electronic warfare doctrines while pointing to ongoing global conflicts as proof of changing combat dynamics. These directives coincide with new domestically engineered software integrations designed to accelerate mission operations and administration.