BackTECH
China's DeepSeek releases long-awaited new AI model
DeepSeek released DeepSeek-V4, touting ultra-long context with a one-million-token capacity and drastically reduced compute and memory costs. The release comes in two versions, Pro and Flash, with a preview open-source version available, building on the company’s 2025 breakthrough with R1 that challenged Western rivals. The move underscores ongoing US-China AI competition and broader industry shifts toward affordable, long-context models.
Why It Matters
The arrival of a cost-efficient, long-context AI model could accelerate mainstream deployment of advanced AI in China and globally, intensifying competition with US firms and prompting broader hardware-cost inflection in the industry.
Timeline
4 Events
April 24, 2026: DeepSeek releases DeepSeek-V4
April 24, 2026
DeepSeek released DeepSeek-V4, which features an ultra-long context of one million tokens and claims world-leading reductions in compute and memory costs. The model is released in two variants: DeepSeek-V4-Pro with 1.6 trillion parameters and DeepSeek-V4-Flash with 284 billion parameters, and is optimized for AI Agent products such as Claude Code, OpenClaw, OpenCode and CodeBuddy. A preview open-source version is also available.
April 23, 2026: White House accuses Chinese entities of AI tech theft
April 23, 2026
The White House accused Chinese entities of an industrial-scale distillation campaign to steal American AI technology, ahead of an expected summit between Donald Trump and Xi Jinping in Beijing next month.
2025: DeepSeek shock and market reaction
2025
Last year, the company’s low-cost R1 model matched top US offerings like ChatGPT while using far less computing power, described as a 'Sputnik moment' that sparked a sell-off in AI-related shares and a broader industry reckoning.
January 2025: DeepSeek debuts R1-powered chatbot
January 2025
DeepSeek burst onto the scene in January 2025 with a generative AI chatbot powered by its R1 reasoning model, challenging assumptions of US AI dominance.