LLM Token Expenditure Index — SDLLMTK
LLM Token Expenditure Index
Daily benchmark for large language model inference token prices across different providers. Published every trading day as ticker SDLLMTK.
ABOUT THIS INDEX
What the LLM Token Expenditure Index tracks
The LLM Token Expenditure Index (SDLLMTK) is the Silicon Data benchmark for large language model inference token pricing. It is published daily as a normalized blended rate expressed in USD per million tokens, drawn from observations across frontier API providers, open-weight inference platforms, brokered dedicated-instance markets, and self-hosted reference deployments.
The index is published as a single blended reading across the LLM inference market, but daily movements carry a clear signature — different segments of the market leave distinct fingerprints on the print, and our daily commentary calls out which corner is driving the move. Every observation entering the calculation is normalized for input/output token mix, context window, batching behavior, and reliability — the index reflects realized cost-to-serve, not headline list pricing on any single provider's pricing page.
Together this represents more than 90% of the addressable global LLM inference spend today.
PRICE DRIVERS
What drives LLM token prices
Provider Pricing Power
The Index captures how model providers translate capability, scarcity, and competitive positioning into market-facing token prices. Rising levels may signal stronger pricing power or sustained willingness to pay for frontier performance.
Market Usage Concentration
The Index is weighted by where usage is concentrated and sustained. When demand shifts toward premium models, expenditure rises; when capable lower-cost models gain adoption, the market becomes more cost-efficient.
Infrastructure-to-Model Efficiency
The Index reflects how GPU costs, memory constraints, inference efficiency, and serving optimization flow through into real token economics. It helps distinguish whether infrastructure improvements are causally contributing to a more efficient token market.
Methodology
How the LLM index is calculated
The LLM Token Expenditure Index is calculated daily from observations across different model providers. Observations are normalized for input/output mix and context window, then filtered for models with sustained usage and market expenditure. Each day's publication is independently validated.
Frequently Asked Questions (FAQ)
The current LLM Token Expenditure Index reads $4.20 per 1,000,000 tokens for the frontier segment and $0.85 per 1,000,000 tokens for the open-weight segment as of Apr 22, 2026, based on the Silicon Data LLM Token Expenditure Index (ticker SDLLMTK). Each reading is a normalized blended rate across major model providers.
Frontier models command premium pricing for state-of-the-art capability, scarce capacity, and strong provider pricing power, while open-weight models can be served on commodity infrastructure by many competing providers. The Index tracks both as separate segments so you can see the spread — and how it narrows or widens as capable lower-cost models gain adoption.
Daily movements reflect three forces: provider pricing power, where inference usage is concentrated across models, and infrastructure-to-model efficiency — GPU costs, memory constraints, and serving optimization. Every observation is normalized for input/output token mix, context window, and batching before it enters the calculation.
It tracks the realized, blended cost-to-serve of LLM inference in USD per million tokens — drawn from frontier API providers, open-weight inference platforms, brokered dedicated-instance markets, and self-hosted reference deployments. It reflects what the market actually pays, not headline list pricing on any single provider's page.
Token prices are downstream of compute economics: GPU rental rates, memory constraints, and serving efficiency all flow into cost-to-serve. The LLM Token Expenditure Index sits alongside Silicon Data's GPU rental indices (SDH100RT, SDA100RT, and others), letting you trace how infrastructure costs translate into token economics.
SDLLMTK is published daily as a normalized blended rate across the 20+ models in the daily basket, weighted by where inference usage is concentrated. Observations are normalized for token mix, context window, batching, and reliability; statistical outliers are removed; and each day's publication is independently validated.
Get access to Silicon Data
Put the LLM Token Expenditure Index to work — track frontier and open-weight token economics daily, alongside Silicon Data's full family of GPU and compute indices.