MWS Cloud Cuts Overnight Neural Network Response Prices by Up to 30% in MWS GPT Model Hub
MWS Cloud has introduced an overnight discount on inference, meaning on each neural network response, in its MWS GPT Model Hub service. From 1:00 to 7:00, prices are up to 30% below the base rate, every day, including holidays. It works like a night electricity tariff: customer chat stays on the daytime rate, while bulk text generation and document analysis can be shifted to the night. You don't need to change your integration, so batch jobs are easy to put on a nighttime schedule.
AI-processed from TG: @ai_machinelearning_big_data (Machinelearning); edited by Hamidun News
Russian cloud provider MWS Cloud has made neural network responses cheaper during night hours. In the MWS GPT Model Hub service, from 1:00 to 7:00, inference, meaning each request to which an already trained model returns an answer, costs up to 30% less than the base price. Think of a night electricity tariff: a washing machine doesn't care when it runs, and you can pay less. The same goes for bulk AI tasks that don't need an instant result.
What exactly was launched
The discount applies every day, including weekends and holidays. The words "up to 30%" matter: the size of the discount may vary, so exact figures for specific models should be checked in the tariffs on the website. The company claims it was the first in Russia to introduce a nighttime inference discount. This is the company's own claim, and we have not verified it independently.
For developers, the mechanism is simple. You pay for each model response, and at night it costs less. According to MWS Cloud, no integration changes are required: the API address and request format stay the same, and only the time you send requests changes.
Which tasks to send at night
The discount is useful where the result is needed not right away but by morning. The company itself cites such examples:
- bulk text generation, for example product descriptions or newsletters;
- analysis of large batches of documents;
- preparation of regular reports;
- processing of customer reviews;
- creation of embeddings, that is, numerical "fingerprints" of texts used for search and RAG.
If you run a customer chatbot, the night tariff won't help: the user expects an answer immediately. But recalculating a knowledge base index or a daily summary of support tickets can well run on a schedule at 2–3 a.m.
The practical takeaway: split your load into "urgent" and "deferred." For the second group, set up a queue or scheduler that launches tasks in the night window. If volumes are large, saving up to 30% on that portion of traffic can become a noticeable line in the budget. Keep in mind the caveat too: speed and availability at night may differ from daytime. The company doesn't write about this, so it's worth testing on a small batch first.
Playground and the model list
Alongside the tariff, the service has added a Playground. It's an interface where you can test models and prompts on your own tasks before connecting them to a product. There you can compare answer quality, processing speed and cost, and choose the right option. For teams choosing between several models, this saves time: you don't need to write a separate script for the first comparison.
Currently available in MWS GPT Model Hub:
- GLM-5.3;
- Kimi K2.6;
- Qwen3.6 35B (A3B);
- Gemma 4 31B;
- gpt-oss-120b.
The catalog includes models for text, code, images, speech and audio, as well as for search and RAG. You connect via an OpenAI-compatible API. If your code already works with the OpenAI client, switching to this service usually means changing the address and key. Requests are processed in the MWS Cloud cloud, and according to the company, data does not leave it. For Russian companies with data storage and processing requirements, this could be a weighty argument.
What this means for the market
The night tariff is a step toward more flexible inference pricing. Other providers have long used similar logic with batch modes: if a customer is willing to wait, they pay less. In Russia this practice is only just emerging, and competitors may respond with their own offers.
For a product manager or founder, the takeaway is this: the cost of AI in a product depends not only on the choice of model but also on when it runs. Before switching to a cheaper model and losing quality, check what share of your tasks can simply be moved to night. Start by running your typical requests in the Playground, calculate daytime and nighttime costs, and decide which processes move to a schedule.
- Nov 21, 2026California Sues OpenAI: Its AI Agents Escaped Control at Least 80 Times
- Nov 21, 2026OpenAI GPT-6 Cyber Found About 35 Flaws in iPhone, Mac and Other Apple Devices
- Nov 21, 202622 Claude Agents Spent Six Weeks on Black Hole Problems — and Solved Two
- Nov 21, 2026Google to split Gemini chat history from model training, so you can opt out without losing your history
Key Questions Answered
What exactly was launched?
The discount applies every day, including weekends and holidays. The words "up to 30%" matter: the size of the discount may vary, so exact figures for specific models should be checked in the tariffs on the website. The company claims it was the first in Russia to introduce a nighttime inference discount. This is the company's own claim, and we have not verified it independently. For developers, the mechanism is simple. You pay for each model response, and at night it costs less.
Which tasks to send at night?
The discount is useful where the result is needed not right away but by morning. The company itself cites such examples: bulk text generation, for example product descriptions or newsletters; analysis of large batches of documents; preparation of regular reports; processing of customer reviews; creation of embeddings, that is, numerical "fingerprints" of texts used for search and RAG. If you run a customer chatbot, the night tariff won't help: the user expects an answer immediately.
What are the key details on Playground and the model list?
Alongside the tariff, the service has added a Playground. It's an interface where you can test models and prompts on your own tasks before connecting them to a product. There you can compare answer quality, processing speed and cost, and choose the right option. For teams choosing between several models, this saves time: you don't need to write a separate script for the first comparison. Currently available in MWS GPT Model Hub: GLM-5.3; Kimi K2.6; Qwen3.6 35B (A3B); Gemma 4 31B; gpt-oss-120b.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.