Service limits for Azure AI Translator Service
This article provides both a quick reference and detailed description of Azure AI Translator Service character and array limits for text and document translation.
Text translation
Charges are incurred based on character count, not request frequency. Character limits are subscription-based.
Character and array limits per request
Each translate request is limited to 50,000 characters, across all the target languages. For example, sending a translate request of 3,000 characters to translate to three different languages results in a request size of 3,000 × 3 = 9,000 characters and meets the request limit. You're charged per character, not by the number of requests, therefore, we recommend that you send shorter requests.
The following table lists array element and character limits for each text translation operation.
Operation | Maximum Size of Array Element | Maximum Number of Array Elements | Maximum Request Size (characters) |
---|---|---|---|
Translate | 50,000 | 1,000 | 50,000 |
Transliterate | 5,000 | 10 | 5,000 |
Detect | 50,000 | 100 | 50,000 |
BreakSentence | 50,000 | 100 | 50,000 |
Dictionary Lookup | 100 | 10 | 1,000 |
Dictionary Examples | 100 for text and 100 for translation (200 total) | 10 | 2,000 |
Character limits per hour
Your character limit per hour is based on your Translator subscription tier.
The hourly quota should be consumed evenly throughout the hour. For example, at the F0 tier limit of 2 million characters per hour, characters should be consumed no faster than roughly 33,300 characters per minute. The sliding window range is 2 million characters divided by 60 minutes.
You're likely to receive an out-of-quota response under the following circumstances:
- You reached or surpassed the quota limit.
- You sent a large portion of the quota in too short a period of time.
There are no limits on concurrent requests.
Tier | Character limit |
---|---|
F0 | 2 million characters per hour |
S1 | 40 million characters per hour |
S2 / C2 | 40 million characters per hour |
S3 / C3 | 120 million characters per hour |
S4 / C4 | 200 million characters per hour |
Limits for multi-service subscriptions are the same as the S1 tier.
These limits are restricted to Microsoft's standard translation models. Custom translation models that use Custom Translator are limited to 3,600 characters per second, per model.
Latency
The Translator has a maximum latency of 15 seconds using standard models and 120 seconds when using custom models. Typically, responses for text within 100 characters are returned in 150 milliseconds to 300 milliseconds. The custom translator models have similar latency characteristics on sustained request rate and can have a higher latency when your request rate is intermittent. Response times vary based on the size of the request and language pair. If you don't receive a translation or an error response within that time frame, check your code, your network connection, and retry.
Document Translation
Note
- Document Translation doesn't support translating secured documents such as those with an encrypted password or with restricted access to copy content.
- When translating documents with content in multiple languages (batch operations only), the feature is intended for complete sentences in a single language. If sentences are composed of more than one language, the content may not all translate into the target language.
Asynchronous (batch) operation limits
Attribute | Limit |
---|---|
Document size | ≤ 40 MB |
Total number of files. | ≤ 1000 |
Total content size in a batch | ≤ 250 MB |
Number of target languages in a batch | ≤ 10 |
Size of Translation memory file | ≤ 10 MB |
Synchronous operation limits
Attribute | Limit |
---|---|
Document size | ≤ 10 MB |
Total number of files. | 1 |
Total number of target languages | 1 |
Size of Translation memory file | ≤ 1 MB |
Translated character limit | 6 million characters per minute (cpm). |
Next steps
Feedback
https://aka.ms/ContentUserFeedback.
Coming soon: Throughout 2024 we will be phasing out GitHub Issues as the feedback mechanism for content and replacing it with a new feedback system. For more information see:Submit and view feedback for