计费说明
5 条价格记录IBM's official watsonx.ai pricing page is a use-case-based pricing panel: figures are quoted inside official worked examples rather than a single unified rate card, and they sit on top of a monthly standing charge. Inference is billed per 1M tokens; fine-tuning per GPU-hour; model hosting per hour; some services are billed in Capacity-Unit-Hours ($0.42/CUH).
价格明细
不同模态使用真实计费单位| 模型 / 项目 | 类型 | 计费单位 | 价格 | 输入 | 输出 | 缓存命中 | 币种 | 人民币参考 | 备注 |
|---|---|---|---|---|---|---|---|---|---|
| Granite-3-2b-instruct | Token | per 1M tokens | — | 0.1USD | — | — | USD |
输入 ¥0.72 |
Inference, per 1M tokens (page does not split input/output) |
| Granite-3-3-8b-instruct | Token | per 1M tokens | — | 0.2USD | — | — | USD |
输入 ¥1.44 |
Inference, per 1M tokens |
| Embeddings (watsonx.ai) | Token | per 1M tokens | — | 0.1USD | — | — | USD |
输入 ¥0.72 |
'Embeddings: $0.10/M tokens' as printed on the official page |
| Granite-8B hosting | Token | per hour | — | 5.22USD | — | — | USD |
输入 ¥37.584 |
Dedicated hosting, example: 720h x $5.22/h |
| LoRA/QLoRA fine-tuning (A100) | Token | per GPU-hour | — | 5.88USD | — | — | USD |
输入 ¥42.336 |
H100 option $13.86/h |
人民币一列为按 1 USD ≈ 7.20 CNY 参考汇率折算，仅供参考，不是厂商官方人民币价格。
来源与核验
legacy-unreviewed迁移保留旧内容；来源是旧链接线索，本次未重新打开核验。