ANTILATTICE EDGE · ON-DEVICE INTELLIGENCE
The fastest AI ever built for the device in your pocket.
A 1B fp16 model. 256K context in 4GB of RAM. No cloud.
Meet Antilattice Edge. Built to live in your pocket — and nowhere else.
Trace how frontier capability fits in 4GB of RAM — a full fp16 model whose cache grows with meaning, not with every pair.
01 / The memory math
256K context. 4GB of RAM. It fits.
A 1B fp16 model is 2GB of weights. Dense attention adds ~192KB of KV per token — 51.5GB at full context, thirteen times the entire budget. Latent Graph Attention routes the cache along learned edges at ~2KB per token, and the whole footprint lands at 3.1GB.
The cache alone outweighs the phone thirteen times over. Squeezed into one budget, dense holds roughly 8K tokens — Edge holds 256K, a 32× larger window.
02 / Measured on-device
78% of short-context speed — at 256K.
Internal benchmarking on a 2026 flagship smartphone with a 4GB memory budget, normalized to short-context generation speed. Dense fp16 KV exhausts the budget before 16K tokens and is memory-infeasible from 32K; Antilattice Edge keeps serving to 256K.
78%
of short-context generation speed, held at 256K tokens
32×
more context than dense fp16 fits in the same 4GB budget
03 / Built for
The workloads that can't afford the cloud.
04 / The memory wall
Local AI was never slow — it was starved
Figures
Fig. 01 — the cache that never stops
Memory is the wall
Dense KV grows at ~192KB per token for a 1B model. At 256K that's 51GB of cache — before weights, before activations, before the OS gets a byte.
Fig. 02 — the usual escape hatch
Quantization hides the symptom
int4 weights shrink the model but not the dense cache. Context stays capped, precision gets thrown away, and the wall stays where it was.
Fig. 03 — what Edge keeps
Route the cache itself
Latent Graph Attention keeps KV only along learned edges — ~2KB per token. The cache collapses, fp16 survives, and 256K fits in 4GB.
05 / Get started
Same intelligence.
No data center.
We build frontier models that live entirely on the device. If your product needs intelligence that never leaves the hardware, we want to hear from you.
> hello@antilattice.com