The Hologres long memory service uses a pay-as-you-go (postpaid) billing model. The system meters usage for each billable item on an hourly basis and generates bills accordingly.
-
The Hologres long memory service is in free public preview. The public preview ends at 10:00:00 (UTC+8) on August 1, 2026. Standard charges apply after this date.
-
During the public preview, you still receive postpaid bills from the Hologres long memory service. These bills show the usage of compute, storage, and model billable items as usual, but the actual charges are USD 0. We recommend that you monitor your usage to estimate costs after the preview ends.
Billable items
The Hologres long memory service includes the following billable items:
|
Billable item |
Description |
Billing method |
|
Compute |
Resources consumed by API calls for memory creation, modification, deletion, and query operations. Maximum supported rate: 20 QPS. |
Billed by the number of API calls |
|
Storage |
Storage resources for memory embeddings and related data. |
Billed by the number of memory entries |
|
Model |
Resources consumed by model operations: memory extraction, memory vectorization, and recall result reranking. |
Billed by actual token consumption |
The following models are used for model-related billing:
|
Purpose |
Model |
|
Text generation |
qwen3.7-plus |
|
Text embedding |
text-embedding-v4 |
|
Text reranking |
qwen3-vl-rerank |
Pricing
-
Compute (API calls): USD 0.000537313/call
-
Storage (memory entries): USD 0.00223881/10,000 entries/hour
-
Model: Pricing is as follows
Model
Input tokens per request
Input price (USD/million tokens)
Output price (USD/million tokens)
qwen3.7-plus
0 < tokens ≤ 256K
0.3312
1.3212
qwen3.7-plus
256K < tokens ≤ 1M
0.9912
3.9612
text-embedding-v4
N/A
0.0864
N/A
qwen3-vl-rerank
N/A
0.12
N/A