Silicon Data token index
SDLLMTK, the Silicon Data LLM Token Expenditure Index: what a million tokens of language-model inference costs across labs, weighted by spend, and why its headline can fall while every price rises.
What it measures#
The Silicon Data LLM Token Expenditure Index, ticker SDLLMTK: what a million tokens of language-model inference costs to buy, in US dollars, blended across labs and weighted by how much is spent on each. It is the demand side of GPU rental, where the rental price is the supply side, and the one series on Market Data that prices what the GPUs produce rather than the GPUs.
Silicon Data build it from two halves, an open-weight index and a proprietary index, and publish the blend. Only the blend has a public source; the two halves came from the authenticated API during a trial and are frozen where they stopped.
The headline can fall while every price rises. Over the 273 days to 30 August 2026 the open-weight half rose 77 percent and the proprietary half rose 82 percent, and the blend fell eight percent, because the weight moved: open-weight models went from 16 percent of token spend to 72 percent. A falling index here is a mix shift toward cheaper models, not cheaper inference, and the blend alone would tell a reader the opposite of what happened.
How it is built#
By Silicon Data, as a spend-weighted blend of two halves they also publish: an index of what a million tokens costs on open-weight models and one for closed frontier models. The blend's weight is the share of token spend going to each half, and it is that weight, not the prices, that has moved the headline most. The site can show only the blend, because only the blend has a public source.
Publication#
- Source
- Silicon Data
- Unit
- USD / M tokens
- Frequency
- Daily
- Ticker
- SDLLMTK
What GPUQuant shows#
The newest 7 days, and nothing older, under a written display licence from Silicon Data that covers display on this site and excludes redistribution and the accumulation of history. The figures are read from Silicon Data's public chart, which serves exactly that window, and the explorer opens the row on its seven-day view.
On Market Data as Blended · Silicon Data under All labs, and as Open LLM · Silicon Data under All labs, and as Proprietary LLM · Silicon Data under All labs.
Reading it beside the others#
- Ornn token index, OpenAI: The transacted price of a million tokens on one lab’s models, which is one of the prices this blend is made of.
- Ornn token index, Anthropic: The same for Anthropic, to read whether a move in the blend is a price change or a shift in which lab’s tokens are bought.
- H100 reference price: The supply side: what the hour that produces the tokens costs.
Sources#
- Silicon Data's public chart, which the figures are read from.
- Silicon Data indices, the RAM and token indices.