The same workload, the same rules, the same legality check. Same orchestrator. Different model. See what the substrate extracts from a 30B model that the brain-locked design cannot reach at any price.
Each turn in an Autogenesis multiplayer game. Cost mid blended between Qwen Flex (main/host pipes) and Qwen Standard (embedded reasoning). The substrate's structured output, segmented memory, and transformation functions keep every turn at minimum token count. The model only emits and consumes what the work needs.
A 4-player multiplayer game runs 25 rounds, 4 player turns + ~3 NPC turns per round. The substrate handles routing, memory, and verification. The model only emits the predictions. Toggle the mode to see 1v1 (2×) and 1v3 (4×) scaling.
A $25/month subscriber paying for Unlimited multiplayer games. Inference budget at 80% margin = $20. The chart shows how many games per month each architecture can afford at that tier.
| Architecture | Per-game cost | Games / month | Headroom |
|---|---|---|---|
| Substrate + Qwen 30B | — | — | VIABLE |
| Brain-locked + Opus 4.6 | — | — | UNVIABLE |
The brain-locked design loses money on inference alone. The substrate design returns 80% margin for the operator before any other cost.
A 30B model with the substrate delivers the same classification rate as Opus 4.6 at roughly 1/30th the per-token cost. The substrate does not make the model smarter. The substrate extracts the capability the model already has by routing everything else to infrastructure.
Same legality check, same rule book, same transformation function. Qwen 30B-A3B through the substrate matches Opus 4.6's effective classification rate per the Open-Autogenesis trace evidence.
The $25 tier is a substrate outcome. The $50, $75, $100 tiers are profitable. The framework-as-brain competitors cannot reach the consumer market segment at any model tier.