Substrate Extraction — Cost Per Game

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.

1. Per-turn cost basis — what each turn actually costs

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.

Human turn
971K input + 50K output
AI player turn
1.14M input + 47K output
NPC turn
149K input + 20K output
Tokens per turn (avg)
across all turn types
Qwen 30B (substrate) Opus 4.6 (brain-locked)

2. Cost per game — the bill that hits the operator

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.

Qwen 30B per user
at selected mode
Opus 4.6 per user
framework-as-brain, same workload
Substrate savings
dollars per user per game
Multiplier
Opus ÷ Qwen
Qwen 30B via substrate Opus 4.6 brain-locked

3. The $25 tier — viable or not?

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.

Side-by-side at the $25 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.

What the substrate extracts

Per-token cost: 1/30th

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.

Classification parity: 99%

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.

Market segment: Opens

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.