44 / Foundation Models & Science
SLM360: On-Device Language Model Under 50MB
We built India's first sub-50MB language model that runs entirely on the device, answering in 39 milliseconds with no server and no connection, private by default because nothing ever leaves the phone.
- Built by
- 360 Labs (Internal)
- Stack & approach
- On-DeviceSub-50MB39ms LatencyDistillationFully Offline
The problem
Language models are too large for deployment on constrained devices, and server round-trips break the offline, private-by-default posture that on-device AI is supposed to provide.
Solution & architecture
A sub-50MB model that runs entirely on-device, answering at 39ms with no connectivity and no server. Quantised and distilled for device-class hardware, SLM360 delivers interactive responses fully offline: no data ever leaves the device.
Research highlights
Nothing leaves the device
No server, no connectivity in the path
Private by default
On-device inference is inherently private
India's first
Sub-50MB on-device language model
