Legal
A 4B legal agent outscored GPT-5.
On this legal benchmark, the 4B specialist scored 78.02% vs. GPT-5 at 75.34%.
A task-specific result—not an overall model ranking.
Read the Typhoon-S paperOn-device AIShopping Agents
Shoppers describe what they need. Edge0 searches approved product data, applies constraints, and prepares a cart for confirmation inside the retailer app.
Preparing the Cartside retail workspace…
01 / ON-DEVICE SHOPPING
Supported shopping tasks Handled on the phone
Edge0 uses a 4-bit, 2B-class retail model for supported catalog search, product comparison, and cart preparation on the phone. Swift reflects the current iPhone proof; Android and WebGPU remain illustrative previews.
import Edge0Kitlet cartside = try await Edge0Agent.configure( local: "edge0/retail-2b-int4", fallback: .cloudFallback, tools: [search_products, get_promotions, get_basket, propose_items, propose_basket_change], confirmation: .required)for try await event in cartside.stream(shopperMessage) { render(event.route, event.fallbackReason, event.content)}Hi — I'm Cartside inside AmazingFood. I can answer product questions, compare options, and prepare a basket. I always ask before changing it.
iOSCurrent native proof · iPhone 15 Pro Max
Illustrative integration code. The current device proof is iPhone-only; Android and WebGPU are unverified previews.
02 / DEPLOYMENT & ROUTING
From retailer data to a cart ready for confirmation
Connect approved retail data. Test real shopping tasks. Route supported work on-device. Verify the outcome before release.
- 1
Connect retailer systems
Connect approved catalog, price, promotion, cart, and policy data.
- 2
Test shopping tasks
Replay product questions, comparisons, cart proposals, and confirmation boundaries.
- 3
Route each request
Keep supported work on-device, send harder cases to cloud fallback, and hand sensitive work to the retailer team.
- 4
Verify the outcome
Check the proposed cart against shopper constraints and attach quality gates to every release.
Grow Your AgentNot Your AI Bill
Edge0 is designed to reduce round trips and make cloud usage more predictable while keeping fallback available for harder requests.
Illustrative inference cost as workload scales
Cloud-only inference uses cloud capacity for every request. Hybrid local and cloud-fallback routing can complete supported requests locally. This is an illustrative comparison, not a pricing or savings claim.
Illustrative inference cost
Workload scale
- 0 usersBaseline
- PilotFirst task
- LearnMeasure route mix
- ExpandMore shoppers
- ScaleHigher volume
03 / THE DEVICE CURVE
Small, specific agents are entering their era
Edge0’s current proof runs a 4-bit, 2B-class retail model on an iPhone. Improving device memory and compute can move more supported work closer to the shopper; broader platform support remains subject to testing.
Smartphone price falls as memory capacity rises.
Illustrative annual scenarios through 2024, with projections from 2025 to 2030.
Average pricePrice projectionAverage RAMRAM projection
Illustrative smartphone scenarios show average price declining from 470 dollars in 2010 to 195 dollars in 2030, while average RAM rises from 0.5 gigabytes to 16 gigabytes. Values from 2025 onward are projections.
04 / RESEARCH EVIDENCE
Small can win when the task is specific
Two focused research benchmarks (not Edge0 retail results).
Medical
A 4B medical ranking model beat Gemini 2.5 Pro.
On ranked-list questions, the 4B specialist scored 94.73% vs. Gemini 2.5 Pro at 68.46%.
Research only—not for diagnosis or clinical decisions.
Read the OpenTyphoon medical research
ONE SHOPPING TASK · ONE MEASURABLE OUTCOME