DeepSeek Just Dropped V4. The AI Race Between China and the US Just Got Louder.
Quick Reads
- DeepSeek released a preview of its V4 large language model on April 24, 2026.
- V4 comes in two versions: V4-Pro with 1.6 trillion parameters and V4-Flash with 284 billion, both with a 1 million token context window, up from 128,000 in V3.
- The model runs on Huawei Ascend chips and Chinese-made silicon rather than Nvidia hardware, a direct challenge to the US export control strategy.
- Both versions are open source and already rank as the top open-weight model on the Vibe Code Benchmark.
A year ago, DeepSeek’s R1 model arrived and sent investor portfolios into freefall. NVIDIA lost hundreds of billions in market cap in a single session. Data centre spending plans were questioned. The assumption that only American labs with access to the best American chips could build frontier AI took a public hit it has not fully recovered from. Now DeepSeek is back with V4, and while it is unlikely to produce the same market shock, the technology and the geopolitical story behind it are arguably more significant than the original.
Chinese artificial intelligence startup DeepSeek released a preview version of its long-awaited V4 large language model on Friday, allowing users to test its new capabilities and features. Reported that the release follows more than a year of anticipation, with some analysts having expected the model to arrive as far back as the Lunar New Year in early 2026.
DeepSeek-V4 features an ultra-long context of one million words. V4-Pro has 1.6 trillion parameters while the V4-Flash has 284 billion, which refines the model’s decision-making ability. How-To Geek That context window jump, from 128,000 tokens in V3 to 1 million in V4, is not a cosmetic upgrade. It means the model can process and reason across far longer documents, conversations, and codebases in a single session, which directly expands what it is useful for in agent workflows.
DeepSeek says the new V4 open-source models have big improvements in knowledge, reasoning and in their agentic capabilities, the ability to perform complex tasks and workflows autonomously. The company’s own benchmarking claims place V4-Pro ahead of all rival open models for maths and coding, with performance only marginally below Google’s closed Gemini 3.1-Pro and OpenAI’s GPT-5.4. Independent evaluation platform Vals AI assessed V4 as the number one open-weight model on its Vibe Code Benchmark.
The chip story is the one that matters most beyond the benchmark tables. Chinese tech giant Huawei confirmed that its latest AI computing cluster, powered by its Ascend AI processors, can support DeepSeek’s V4 model. Wei Sun of Counterpoint Research highlighted the fact that V4 is run on domestic chips from Huawei and Cambricon, in comparison to R1, which was trained on Nvidia hardware. He said this allows AI systems to be built and deployed without relying solely on Nvidia, which is why V4 could ultimately have an even bigger impact than R1, accelerating adoption domestically and contributing to faster global AI development overall.
That is a direct response to years of US export controls designed to slow China’s AI development by limiting access to Nvidia’s most advanced processors. If V4 performs at or near the frontier level while running on domestic silicon, the strategic calculation behind those export controls becomes significantly harder to defend. Chinese AI firms have effectively closed the AI performance gap with their US rivals, according to the Stanford AI Index 2026.
Shares of Chinese contract chip manufacturers rose sharply in Hong Kong after the V4 announcement, with SMIC surging 9 per cent and Hua Hong Semiconductor surging 15 per cent. The market read the chip story the same way the analysts did: a competitive Chinese AI model running natively on domestic hardware is a different kind of problem than one dependent on smuggled Nvidia chips.
V4’s debut is unlikely to have the same market impact as R1, because traders have already priced in the reality that Chinese AI is competitive and cheaper to use. But for anyone building on AI, advising on it, regulating it, or competing with it, DeepSeek V4 is not a story about benchmarks. It is a story about how fast the gap between American AI leadership and Chinese AI capability is closing, and what that means for policy, investment, and technology strategy over the next several years.
