Global06 Model serving
Open-model price basket
How far below the official API price the cheapest third-party provider sells the same open model. A fixed basket, recorded weekly.
Lowest third-party price ÷ official price (median)27%For the same open model, third-party inference providers charge about a quarter of the official price.19%–44%
Dataset
| Date | Model | Official | Lowest | Lowest-price provider | Ratio | Status |
|---|---|---|---|---|---|---|
| 2026-11-03 | Kimi K3 | 6.00 | 2.61 | Wafer · US | 44% | Market report |
| 2026-11-03 | MiniMax M3.1 Flash | 0.263 | 0.087 | Tenstorrent · US | 33% | Market report |
| 2026-11-03 | DeepSeek V4.1 Flash | 0.525 | 0.113 | DeepInfra · US | 22% | Unverified |
| 2026-11-03 | GLM-5.3 | 2.15 | 0.402 | Wafer · US | 19% | Market report |
Unverified · DeepSeek V4.1 Flash：$0.113 at a 3:1 input/output blend equals the previous V4 Flash price ($0.09 / $0.18); the real V4.1 Flash price may be higher. It will be updated once checked, with a changelog entry.
METHODOLOGY · Methodology
- Blended price: input and output prices weighted by usage.
- Compare within a model only: tokenizers differ, so token counts are not comparable across models.
- Official prices come from each model maker’s pricing page; the lowest third-party price comes from the OpenRouter public API and provider sites on the same date (once automated collection is live). This issue comes from a 2026-11-03 market report and is being checked line by line.
Changelog
2026-11-03First issue, four models
HistoryBuilding history: one point a week; the trend appears after eight weeks.
How to cite
Watts / FLOPs (2026). Open-model price basket, 2026-11-03. wattsandflops.com/en/data/open-model-basket/