Turbo Cloud reports first results of new AI models: million-token context and 753 billion parameters

Russian cloud provider Turbo Cloud, part of the RTK-DATA group, has expanded its Foundation Model Hub with top new-generation AI models. A week after the models DeepSeek-V4-Flash-0731 and GLM 5.2 appeared on the platform, the company observes steady interest from developers and business users. DeepSeek-V4-Flash-0731 excels in resource-intensive tasks requiring speed, accuracy, and the ability to hold a giant context: its million-token context allows analyzing large documentation arrays, code repositories, or heterogeneous knowledge bases in a single pass, while the updated architecture improves text handling and code generation, including multi-step agentic scenarios. GLM 5.2 targets complex engineering and analytical workloads, from multi-step reasoning and calculations to building autonomous agent systems; open weights make it sought after in projects requiring deep customization and audit. The catalog also includes two Qwen models: Qwen3.6-35B-A3B with 35 billion parameters and a context of up to 262 thousand tokens, working with text and images for typical business tasks like content generation, search across internal bases, and routine automation, and Qwen3.5-122B-A10B with 122 billion parameters for heavier corporate scenarios such as processing large documentation volumes, analytics, and advanced AI assistants. All models are provided via a unified OpenAI-compatible API, allowing switching without code modifications, integration does not require deploying own GPU infrastructure, and flexible key management and configurable consumption limits (daily, weekly, monthly) let teams control resource usage. Payment follows a pay-as-you-go model, charging only for actually processed tokens.