Log scale — bar length is logarithmic: each grid step is ×10. Two bars that look “a bit” apart can differ by an order of magnitude. Don’t read it linearly.
Price Space
Each dot is a model (log–log scale — every grid step is ×10). Dashed line: output = input. Dots above the line charge more for output than input. A tail stretching left from a dot marks its cached-input price — the longer the tail, the deeper the cache discount; no tail means no cached price published.
Full data
| Model | Input | Cached input | Output | Source |
|---|
Methodology
- Reading the chart: bars diverge from the center column — left is input price, right is output price, both on logarithmic scales. The cached-input price is embedded inside the input bar as the solid segment.
- Prices are official long-term stable prices collected from each provider's pricing page — short-context tier where tiered, off-peak where peak/off-peak pricing exists; permanent official discounts are adopted, time-limited promotions ignored.
- CNY-priced models are converted to USD at the Bank of China mid-rate on the collection date (rate recorded per model).
- «—» means the provider publishes no cached-input price for that model.
- Prices are refreshed automatically by an Agent on a schedule, following a documented collection protocol — the collection date above reflects the most recent run.