64% build savings.
Zero compromise on the floor.
Factory codebases are 55% boilerplate. Tiered routing routes it to Flash-Lite. Safety interlocks and SCADA integration stay on Opus. Right model, right task — across the full production lifecycle.
Claude Opus 4.8
Orchestrator — safety & quality decisions
Gemini 3.5 Flash
Executor — sensor analysis & diagnostics
Gemini 3.1 Flash-Lite
Routine — status lookups & boilerplate
AA scores from Artificial Analysis Intelligence Index (Jun 2026). Higher = stronger reasoning.
Industrial IoT Platform SDLC
Factory codebases are 55% boilerplate — unit tests, CRUD endpoints, Dockerfiles. Flash-Lite handles those at $0.025/task. Opus handles safety interlocks and SCADA integration.
Design safety interlock logic for press line
Hard real-time constraints, IEC 61508 SIL-2 compliance; a logic error can cause machine damage or injury.
Build SCADA/PLC integration layer (OPC-UA gateway)
Bridging OPC-UA, Modbus, and MQTT across heterogeneous PLCs requires deep protocol reasoning and fault tolerance.
Implement vibration anomaly detection model
Signal processing pipeline with domain-specific thresholds; false positives cause costly unplanned downtime.
Architect multi-site historian data replication
Consistency across edge historians and cloud time-series requires careful CAP theorem tradeoffs under network partition.
Design ISA-95 production scheduling integration
Aligning MES scheduling APIs with ERP order data across shift boundaries is complex, stateful orchestration.
Build MQTT broker adapter for Siemens S7-1500
Adapting a specific vendor protocol — well-documented but requires precise register mapping and error handling.
Build Modbus RTU polling loop for legacy sensors
Mechanical protocol implementation from PLC documentation — pattern-based but needs accuracy.
Create Grafana dashboard panels for OEE metrics
Translating production KPI specs into PromQL/InfluxQL queries — systematic but requires attention to aggregation.
Write E2E tests for alarm escalation workflow
Multi-step UI interaction tests against well-defined alarm state machine.
Generate IaC templates for edge K3s deployment
Kubernetes manifests from architecture specs — templated, pattern-driven output.
Build REST API for shift handover notes
Standard CRUD with JSON schema validation — well-defined resource structure.
Write OpenAPI specs for all historian endpoints
Extracting types from existing Go handlers — documentation extraction task.
Integrate PagerDuty for critical alarm routing
Webhook configuration following established patterns in vendor docs.
Build CSV export for production batch reports
Deterministic formatting from a defined schema — pure data transformation.
CRUD for work-order management
Standard database operations on well-defined entity schema.
Generate unit tests for 40 sensor parsing fns
Repetitive test generation from clear function signatures.
Seed fixture data for 50 machine profiles
Synthetic test data generation following defined schema.
Add form validation to shift-entry screens
Repetitive Zod schema generation from TypeScript types.
Write Storybook stories for factory UI kit
Boilerplate generation from existing component props.
Write Dockerfile for offline edge deployment
Standard container config for air-gapped factory networks.
Generate Swagger docs for SCADA REST layer
Documentation extraction — pure structural task.
Create DB migration for alarm history table
Schema migration from TypeScript entity definition.
Write address-validation for plant-floor assets
Field format validation logic from spec — mechanical.
Generate email templates for maintenance alerts
HTML email templating from brand guidelines — pattern-driven.
Developers
14 devsSprints
13 sprintsTotal Project Cost Comparison
Selected: Gemini Tiered
Floor Operator Assistant
65% of operator queries are routine status lookups (Flash-Lite). Fault diagnostics go to Flash. Safety overrides and e-stop resets go to Opus — where the wrong answer has OSHA implications.
Select a floor query
Line 3 shift production count: 1,847 units. Target: 2,000. Current OEE: 88.4%.
Line 3 — Current shift: 1,847 units produced (92.4% of 2,000 unit target). OEE: 88.4%. Time remaining: 2h 14min. On track to miss target by ~42 units at current pace.
Line 3 — Current shift: 1,847 units (92.4% of 2,000 target). OEE: 88.4% (Availability 94%, Performance 96%, Quality 98%). At current throughput of 8.3 units/min, you'll finish at 1,958 units — 42 short. Would you like me to flag a pace alert for the supervisor?
Queries / Month
3.0MPer-Query Cost (×1000 = $/K queries)
Annual Savings vs All-Opus
$563K
78% reduction · 65% of queries on Flash-Lite at $0.0011/query
Visual Quality Inspection
Flash scans 5 million parts per year in under a second each. Opus makes the pass/rework/scrap call. Pure Flash accuracy isn't good enough when a wrong disposition scraps a $400 casting.
Select a part type
| Check | Result | Signal |
|---|---|---|
| Surface finish | 3 micro-scratches detected | ↑ Above threshold |
| Edge burrs | 2 locations, 0.3mm height | ↑ Exceeds 0.2mm spec |
| Hole diameter #1 | 12.02mm (spec: 12.00±0.05) | → In spec |
| Flatness | 0.08mm deviation | → Within 0.10mm spec |
| Disposition | REWORK — deburring required | ↑ Flag |
Inspection Pipeline Race
Inspections / Year
5.0MPure Opus
$638K
Hybrid
$458K
Pure Flash
$202K
Hybrid vs Pure Opus
$179K saved/yr
28% reduction at 5.0M parts/year · same Opus disposition quality
Nightly Predictive Maintenance Agent
Reviews 3,000 monitored machines per night — pulling vibration FFTs, thermal snapshots, and lubrication logs. 13 data-fetch steps on Flash; priority triage and supervisor dispatch on Opus.
Planner — Opus
Executors (13) — Gemini Flash
Reviewer — Opus
Click any node to inspect token counts and cost
Cost per machine run
Machines / Night
3KAnnual Cost (365 nights)
Annual Savings
$427K
49% reduction · Opus on 2 of 15 steps only
Where model economics become business outcomes
Six routing strategies applied to the same FactoryOS workload. Cost is calculated from existing token volumes; quality, latency, governance, and value remain inspectable.
Cheapest everywhere is quality-infeasible. Best everywhere wastes spend. Workload-aware routing reaches the efficient frontier; GEAP pushes it further when governance and business value enter the decision.
Lifecycle Summary
FactoryOS savings across Build → Run → Extend, at your current slider settings.
Build Savings (one-time)
All-Frontier
$0
Tiered
$0
Run Savings (annual)
All-Frontier
$0
Tiered
$0
Extend Savings (annual)
All-Frontier
$0
Tiered
$0
All-Frontier vs Tiered — By Phase
Total Annual Savings
$0
Interact with the scenarios above to see your savings