Compute scarcity
Can raise TPI
GPU bottlenecks, memory-heavy inference, and high frontier-model demand can make inference more expensive, pushing TPI up and TPX down.
TPX detail
US Token Purchasing Index
Token purchasing power for the same reviewed basket, measured in mTPD: million tokens per USD. Higher TPX means one dollar buys more AI inference tokens.
TPX mTPD
0.16
TPI price
$8.73
24h TPX change
0.00%
Last updated
May 18, 2026, 11:01 AM
TPX Historical Chart
Early history is modeled from price-impact factors until enough real daily source snapshots exist.
Price Impact Factors
The core driver is compute affordability. Technology progress and supply expansion tend to push TPI down and TPX up. Compute scarcity or frontier-demand congestion can push TPI up and TPX down.
Technology progress
Can lower TPI
Better chips, serving efficiency, distillation, caching, and competition can make inference cheaper, pushing TPI down and TPX up.
Frontier demand mix
Can raise TPI
More traffic shifting to premium reasoning, coding, and agent workloads can raise the weighted token price even when base compute gets cheaper.
Provider share rebalance
Mixed effect
Daily market-share proxy updates can move weight toward cheaper or more expensive providers.
TPX Formula
TPX is calculated as weighted average million tokens per USD. It uses the same provider basket and weighting method as TPI, then expresses the result as purchasing power.
How TPX Moves
TPX rises when AI inference becomes cheaper. Technology progress, better chips, serving efficiency, and competition tend to push TPI down and TPX up. Compute scarcity, GPU bottlenecks, or heavy demand for frontier models can push TPI up and TPX down.
Basket Constituents
Constituents and price sources are disclosed for transparency. Exact live constituent weights are calculated from market-share, revenue, and adoption signals and will be summarized publicly at an appropriate reporting frequency.
GPT-5.5
OpenAI
United States / frontier
Weighting signal tracked internally
- Input USD/1M
- 5.0000
- Output USD/1M
- 30.0000
- Composite mTPD
- 0.0889
- Cached input
- 0.5000
official source / May 17, 2026, 12:00 AM
Claude Opus 4.7
Anthropic
United States / frontier
Weighting signal tracked internally
- Input USD/1M
- 5.0000
- Output USD/1M
- 25.0000
- Composite mTPD
- 0.1000
- Cached input
- 0.5000
official source / May 17, 2026, 12:00 AM
Gemini 3.1 Pro Preview
United States / frontier
Weighting signal tracked internally
- Input USD/1M
- 2.0000
- Output USD/1M
- 12.0000
- Composite mTPD
- 0.2222
- Cached input
- 0.2000
official source / May 17, 2026, 12:00 AM
Grok 4.3
xAI
United States / frontier
Weighting signal tracked internally
- Input USD/1M
- 1.2500
- Output USD/1M
- 2.5000
- Composite mTPD
- 0.6400
- Cached input
- n/a
official source / May 17, 2026, 12:00 AM
| Provider | Model | Input USD/1M | Output USD/1M | Composite mTPD | Cached | Source |
|---|---|---|---|---|---|---|
| OpenAI | GPT-5.5 United States / frontier Weighting signal tracked internally | 5.0000 | 30.0000 | 0.0889 | 0.5000 | official May 17, 2026, 12:00 AM |
| Anthropic | Claude Opus 4.7 United States / frontier Weighting signal tracked internally | 5.0000 | 25.0000 | 0.1000 | 0.5000 | official May 17, 2026, 12:00 AM |
| Gemini 3.1 Pro Preview United States / frontier Weighting signal tracked internally | 2.0000 | 12.0000 | 0.2222 | 0.2000 | official May 17, 2026, 12:00 AM | |
| xAI | Grok 4.3 United States / frontier Weighting signal tracked internally | 1.2500 | 2.5000 | 0.6400 | n/a | official May 17, 2026, 12:00 AM |