Quotas and limits in Yandex Foundation Models
Written by
Updated at May 13, 2025
Yandex Foundation Models has the following limits:
- Quotas are organizational restrictions that can be changed by technical support on request.
- Limits are technical limitations due to Yandex Cloud architectural features. The limits cannot be changed.
If you need more resources, contact support and tell us which quotas you need to increase and by how much.
Quotas
| Type of limitation | Value |
|---|---|
| Text vectorization | |
| Number of text vectorization requests per second | 10 |
| Text generation | |
| Number of concurrent generations in synchronous mode, YandexGPT Lite model | 10 |
| Number of concurrent generations in synchronous mode, YandexGPT Pro model | 10 |
| Number of concurrent generations, YandexGPT Pro 32k | 1 |
| Number of concurrent generations in synchronous mode, Llama 8b1 model | 10 |
| Number of concurrent generations, Llama 70b model | 1 |
| Number of requests per second, asynchronous mode (request) | 10 |
| Number of requests per second, asynchronous mode (getting a response) | 50 |
| Number of requests per hour, asynchronous mode (request) | 5,000 |
| Number of tokenization requests per second | 50 |
| Model working mode: Batch | |
| Number of runs per hour | 10 |
| Number of runs per day | 100 |
| Text classification | |
| Number of text classification requests per second | 1 |
| Image generation | |
| Number of generation requests per minute | 500 |
| Number of generation requests per day | 5,000 |
| Number of result requests per second | 50 |
| Model tuning | |
| Number of fine-tuning runs per day | 10 |
| Number of fine-tuning runs per hour | 3 |
| Datasets | |
| Number of uploaded datasets | 100 |
| Maximum size of one dataset | 5 GB |
| Total size of datasets | 300 GB |
Limits
| Type of limitation | Value |
|---|---|
| Period to store results of asynchronous requests on the server | 3 days |
| Text vectorization | |
| Number of input tokens | 2,000 |
| Output vector size | 256 |
| Text generation | |
| Maximum number of tokens in response via API | 2,000 |
| Maximum number of tokens per response in the management console | 1,000 |
| Total number of tokens in request and response, 3rd generation models | 8192 |
| Total number of tokens in request and response, 4th generation YandexGPT Lite model | 8192 |
| Total number of tokens in request and response, 5th generation YandexGPT Lite model | 32,000 |
| Total number of tokens in request and response, synchronous mode of the 4th generation YandexGPT Pro model | 8192 |
| Total number of tokens in request and response, asynchronous mode of the 4th generation YandexGPT Pro model | 32,000 |
| Total number of tokens in request and response, YandexGPT Pro 32k model | 32,000 |
| Total number of tokens in request and response, 5th generation YandexGPT Pro model | 32,000 |
| Number of free requests per hour for users without a billing account. Available only in the management console | 10 |
| Assistants | |
| Maximum number of assistants | 1,000 |
| Maximum number of threads | 10,000 |
| Maximum number of users | 10,000 |
| Maximum number of files to upload | 10,000 |
| Maximum file size | 128 MB |
| Number of files per upload | 100 |
| Maximum number of files per search index | 10,000 |
| Maximum number of messages per thread | 100,000 |
| Maximum number of search indexes | 1,000 |
| Maximum number of indexing operations to run | 10 |
| Image generation | |
| Maximum prompt length | 500 characters |
| Number of free requests per minute for users without a billing account. Available only in the management console | 2 |
| Number of free requests per day for users without a billing account. Available only in the management console | 10 |
1 Llama was created by Meta. Meta is designated as an extremist organization and its activities are prohibited in Russia.