Guidelines and Quotas
Many of the Amazon Comprehend quotas shown here can be increased if needed for your applications. For information about service quotas and to request a quota increase, see AWS Service Quotas.
Keep in mind the following information when using Amazon Comprehend.
Supported Regions
For a list of AWS Regions where Amazon Comprehend is available, see AWS Regions and Endpoints in the Amazon Web Services General Reference.
Overall Quotas
All operations except asynchronous operations and topic modeling operations have the following quotas:
| Description | Quota/Guideline |
|---|---|
| Character encoding | UTF-8 |
| Document size (UTF-8 characters) | 5,000 bytes |
Synchronous requests to label documents with PII using the ContainsPiiEntities operation have the following quotas:
| Description | Quota/Guideline |
|---|---|
| Character encoding | UTF-8 |
| Document size (UTF-8 characters) | 50,000 bytes |
Amazon Comprehend may store your content to continuously improve the quality
of its analysis models. See the Amazon Comprehend FAQ
Throttling When Using Single Transactions
You may be able to avoid throttling by using the batch operations instead of the single transaction operations. For more information, see Multiple Document Operations.
Multiple Document Operations
The BatchDetectDominantLanguage, BatchDetectEntities, BatchDetectKeyPhrases, and BatchDetectSentiment operations have the following quotas:
| Description | Quota/Guideline |
|---|---|
| Documents per request | 25 |
If you plan to send more than 20 requests per second, you should
consider using the batch operations. Batch operations enable you to send more documents
in each request which may result in higher throughput. For example, when you use the
DetectDominantLanguage operation, you can send up to 20 documents per
second. However, if you use the BatchRequestDominantLanguage operation, you
can send up to 250 documents per second, but processing speed may be lower. For more
information about throttling quotas see Amazon Comprehend Quotas in the Amazon Web Services General
Reference. For more information about using the multiple document APIs,
see Multiple Document Synchronous Processing.
Asynchronous Operations
Asynchronous batch jobs are run with the StartDominantLanguageDetectionJob, StartEntitiesDetectionJob, StartKeyPhrasesDetectionJob, StartPiiEntitiesDetectionJob, and StartSentimentDetectionJob operations. These jobs have the following limits:
| Description | Quota/Guideline |
|---|---|
| Maximum size of each document in jobs that detect entities, key phrases, PII, and languages | 1 MB |
| Maximum size of each document in jobs that detect sentiment | 5 KB |
| Total size of all files in batch | 5 GB |
| Maximum number of files, one document per file | 1,000,000 |
You should use the asynchronous operations:
-
To analyze more than 25 documents at a time
-
To analyze documents larger than 5,000 bytes for keywords and entities
For more information, see Asynchronous Batch Processing.
Document Classification
Document classifier training jobs started with the CreateDocumentClassifier operation, asynchronous document classification jobs started with the StartDocumentClassificationJob , and synchronous document classification requests started with the ClassifyDocument operation have the following limits:
|
General
|
|
|---|---|
| Description | Quota/Guideline |
| Character encoding | UTF-8 |
| Maximum number of classes (multi-class mode) | 1,000 |
| Maximum number of classes (multi-label mode) | 100 |
| Maximum length of class name | 5,000 characters |
| Minimum number of training documents per class (multi-class mode) | 50 |
| Minimum number of training documents per class, (multi-label mode) | 10 |
| Total size of all files in training job | 5 GB |
| Total size of all files in asynchronous job | 5 GB |
| Maximum file size for one file, one document per file | 10 MB |
| Maximum number of files, one document per file | 1,000,000 |
| Maximum number of lines, one document per line (for all files in request) | 1,000,000 |
| Maximum number of documents per synchronous request | 1 |
| Maximum number of augmented manifest files for training a custom classifier | 5 |
| Maximum number of attribute names for each augmented manifest file | 5 |
| Maximum length of attribute name | 63 characters |
|
Real-time analysis
|
|
|---|---|
| Description | Quota/Guideline |
| Maximum number of inference units per account | 100 |
| Maximum number of inference units per endpoint | 10 |
| Maximum throughput per inference unit (characters) | 100/second |
| Maximum throughput per inference unit (documents) | 2/second |
Language Detection
The BatchDetectDominantLanguage, DetectDominantLanguage operations and asynchronous jobs started with the StartDominantLanguageDetectionJob operation have the following limitations:
-
They don't support phonetic language detection. For example, they will not detect "arigato" as Japanese nor "nihao" as Chinese.
-
They may have trouble distinguishing close language pairs, such as Indonesian and Malay; or Bosnian, Croatian, and Serbian.
-
For best results the input text should be at least 20 characters long.
Events
Events detection jobs created with the StartEventsDetectionJob operation have the following limits:
| Description | Quotas |
|---|---|
| Character encoding | UTF-8 |
| Total size of all files in a job | 50 MB |
| Maximum size of each document in a job | 10 KB |
| Maximum number of files, one document per file | 5,000 |
| Maximum number of lines, one document per line (for all files in request) | 5,000 |
Topic Modeling
Topic detection jobs created with the StartTopicsDetectionJob operation have the following limits:
| Description | Quota/Guideline |
|---|---|
| Character encoding | UTF-8 |
| Maximum number of topics to return | 100 |
| Maximum total size of all files in request | 5 GB |
| Minimum total size of all files in request | 500 bytes |
| Maximum file size for one file, one document per file | 100 MB |
| Maximum number of files, one document per file | 1,000,000 |
| Maximum number of lines, one document per line (for all files in request) | 1,000,000 |
For best results, you should include at least 1,000 input documents.
Entity Recognition
Entity recognizer training jobs started with the CreateEntityRecognizer operation, asynchronous entity recognition jobs started with the StartEntitiesDetectionJob operation, and synchronous entity recognition requests started with the DetectEntities operation have the following limits:
|
General
|
||
|---|---|---|
| Description | Minimum | Maximum |
| Number of entities per model/custom entity recognizer | 1 | 25 |
| Document size (UTF-8) | 1 byte | 1 MB |
| Number of documents | 1,000 | 120,000 |
| Document corpus size (all docs in plain text combined) | 5 KB | 100 MB |
|
Real-time analysis
|
|
|---|---|
| Description | Quota/Guideline |
| Maximum number of inference units per account | 100 |
| Maximum number of inference units per endpoint | 10 |
| Maximum throughput per inference unit (characters) | 100/second |
| Maximum throughput per inference unit (documents) | 2/second |
| Maximum number of documents per synchronous request | 1 |
|
Annotations
|
|
|---|---|
| Description | Quota/Guideline |
| Number of annotations per entity | At least 200 |
|
Entity lists
|
|
|---|---|
| Description | Quota/Guideline |
| Number of items in entity list | 1–1,000,000 |
| Length of individual entry (post-strip) in entry list | 1–5,000 |
| Entity list corpus size (all docs in plain text combined) | 5KB–100 MB |
|
Augmented manifest files
|
|
|---|---|
| Description | Quota/Guideline |
| Maximum number of augmented manifest files for training a custom entity recognizer | 5 |
| Maximum number of attribute names for each augmented manifest file | 5 |
| Maximum length of attribute name | 63 characters |
