evidence · section 02 of 10
What does a token actually cost, and how far does effective cost diverge from the posted price?
What this section answers
A 111x posted price range across comparable models, a 47x spread once cost is adjusted for quality, a non-linear frontier above Intelligence Index 50, a ~10% closed-weight premium, and caching and batch as the two largest non-routing levers.
Boundary
The quality adjustment inherits the index's editorial weighting; version overhauls have moved scores by ~23 points.
Source coordinate
Report heading Inference economics — five verified claims, under What the evidence establishes, in The real cost of AI: August 2026.
The inference-economics notebook has 17 references and 5 verified claims:
C1 (verified, 3 independent grade-A sources): The same API workload costs $0.018 to $2.00 across major LLMs — a 111x price range driven by the 21x blended-price spread in the frontier basket.
C2 (verified, 3 independent grade-A sources): Quality-adjusted cost diverges from token price by 47x across the frontier basket — far more than the 2x the original hypothesis predicted. DeepSeek V4-Flash at $2/index-point vs Claude Fable 5 at $94/index-point.
C3 (verified, 3 independent grade-A sources): The price-per-quality frontier is non-linear: models in the 40-50 Intelligence Index range offer 10-47x better cost-per-index-point than models in the 55-60 range. The marginal cost of capability accelerates above index 50.
C4 (verified, 1 grade-A source): Closed-weight flagships carry a 10% premium over open-weight equivalents in blended price ($4.47 vs $4.08/Mtok), but the premium concentrates at the frontier, not uniformly across tiers.
The verified price range, the quality adjustment, and the two largest non-routing levers.
The per-token baseline the narrow model has to beat in production, not in training.