MWS Cloud Cuts Nighttime LLM Prices to 30% of Base Rate From 1:00 to 7:00
MWS Cloud has introduced a nighttime tariff in MWS GPT Model Hub: from 1:00 to 7:00, using language models costs up to 30% of the base price. The principle works like a two-rate electricity meter: bulk text generation, document parsing, and embeddings (numerical "fingerprints" of text used for search) can be moved to the night. The integration code does not need to change. That makes it cheaper to run batch jobs at night and keep live conversations during the day.
AI-processed from CNews ИИ; edited by Hamidun News
Russian cloud provider MWS Cloud has introduced a nighttime tariff for language models in its MWS GPT Model Hub service. It is a platform where companies take ready-made LLMs via API and plug them into their own products. From 1:00 to 7:00 in the morning, the cost of using the models drops to 30% of base prices, so the same requests cost noticeably less at night.
The logic is the same as with a two-rate electricity meter: you run the washing machine at night because a kilowatt-hour is cheaper. Here, the electricity is GPU compute, and the laundry is tasks whose results nobody needs within the second of a request. If you have this kind of workload, you can cut your token bill with almost no changes.
What Was Actually Launched
MWS GPT Model Hub now offers nighttime prices that apply every day, including weekends and holidays. The window lasts six hours: from 1:00 to 7:00. Along with the tariffs, the service opened the Playground, an interactive environment for testing models and queries.
The provider's wording is "will drop to 30% relative to base prices." The message does not specify concrete price lists or the list of models, nor the time zone of the window. Before building the discount into your budget, check in the contract or price list how the nighttime price is calculated and which models it covers.
What Can Be Sent Overnight
The company advises moving everything that does not require a real-time response to the night:
- bulk text generation, for example product descriptions;
- analysis of document batches;
- preparation of regular reports;
- processing of customer reviews;
- creation of embeddings, numerical "fingerprints" of text on which search and RAG are built.
Such tasks typically run as batches: you queue them in the evening and collect the results by morning. A support chatbot or a developer assistant, by contrast, stays on daytime prices, because a user will not wait until one in the morning.
An important detail: you do not need to change the service integration. According to an MWS Cloud representative, costs can be reduced without changes on the client side. In practice, a task scheduler that launches batch scripts within the right window is enough. If your workload is already organized as a queue, restructuring will take hours, not weeks.
Playground: Try a Model Before Integration
In the Playground, teams run models on their own scenarios, compare response quality, and tune parameters with speed and cost in mind. For example, say you want to automate contract review. In one evening you can check whether a cheaper model handles the job, without spending weeks building a solution that turns out to be unsuitable.
Together, the tools make a workable scheme. First, in the Playground, you choose a sufficiently good and inexpensive model. Then you decide which parts of the workload are urgent and which can wait. After that, you redistribute requests and look at the resulting costs. The head of the AI Platform division at MWS Cloud notes that companies are increasingly moving generative AI from pilots into production, and the cost of processing requests is becoming one of their main criteria.
What Else to Know
MWS GPT Model Hub provides models for text, code, images, speech and audio, as well as for search. The service runs on the MWS Cloud Platform, requests are processed on the cloud's own infrastructure, and data, the provider states, does not leave its perimeter. For Russian companies with strict data storage requirements, this is a weighty argument.
There are also limitations. First, the savings go to those who have a substantial share of deferrable tasks. If nearly all your traffic is live conversations, the nighttime tariff will change little. Second, you will need to calculate the real benefit yourself, based on your own volumes. Third, the six-hour window is not elastic: large batches must be planned so they finish before 7:00. Otherwise some requests may fall onto the daytime tariff.
The practical takeaway is simple: separate batch and interactive LLM tasks. Even without the discount, this architecture simplifies queue and cost management, and with the nighttime tariff it also saves money. Clouds have long sold idle capacity at reduced prices, and now the same approach has reached language model services.
- Nov 21, 2026India: 70,000 AI-Suspect Books Found in Tamil Nadu Government Purchases Worth About 60 Million Rubles
- Nov 21, 2026Canva adds a "turn off AI" toggle and a filter that puts human-made works first in its library
- Nov 21, 2026Claude Code writes and publishes SEO articles three times a day on a schedule
- Nov 21, 2026An agent learned to cheat on its own review, and AI judges catch errors with under 55% accuracy
Key Questions Answered
What Was Actually Launched?
MWS GPT Model Hub now offers nighttime prices that apply every day, including weekends and holidays. The window lasts six hours: from 1:00 to 7:00. Along with the tariffs, the service opened the Playground, an interactive environment for testing models and queries. The provider's wording is "will drop to 30% relative to base prices." The message does not specify concrete price lists or the list of models, nor the time zone of the window.
What Can Be Sent Overnight?
The company advises moving everything that does not require a real-time response to the night: bulk text generation, for example product descriptions; analysis of document batches; preparation of regular reports; processing of customer reviews; creation of embeddings, numerical "fingerprints" of text on which search and RAG are built. Such tasks typically run as batches: you queue them in the evening and collect the results by morning.
What are the key details on Playground: Try a Model Before Integration?
In the Playground, teams run models on their own scenarios, compare response quality, and tune parameters with speed and cost in mind. For example, say you want to automate contract review. In one evening you can check whether a cheaper model handles the job, without spending weeks building a solution that turns out to be unsuitable. Together, the tools make a workable scheme. First, in the Playground, you choose a sufficiently good and inexpensive model.
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.