Chinese start-up DeepSeek is at the center of mounting online speculation as social media buzz grows over the impending release of its next open-source artificial intelligence model, DeepSeek-R2. The company, well-known for its cost-efficient technology, has not officially confirmed details about R2’s launch, but discussions online suggest the new model could introduce significant advances in performance and cost savings, raising anticipation throughout the tech sector against the backdrop of the ongoing US-China tech rivalry.
Interest in DeepSeek surged after the company’s rapid emergence in late 2024 and early 2025, when it introduced two advanced open-source artificial intelligence models, V3 and R1. Both models drew attention for being developed at a fraction of the cost and computing resources compared to those required by global tech giants for similar large language model (LLM) projects. Such LLMs underpin generative artificial intelligence applications like ChatGPT, which have become central to both industry and consumer use cases.
According to recent posts circulating on Chinese stock trading social platforms, the upcoming R2 model is reportedly based on a hybrid mixture-of-experts (MoE) architecture, featuring a massive 1.2 trillion parameters. This architecture divides models into specialized sub-networks that handle different aspects of data processing, resulting in substantially reduced computation needs during pre-training and faster inference times. Notably, R2 is claimed to be up to 97.3 per cent less expensive to build than OpenAI’s GPT-4o. These rumors, if substantiated, could position DeepSeek R2 as a transformative player in global artificial intelligence competition, particularly as Chinese start-ups vie to lessen dependence on Western technologies amid ongoing international tech tensions.
