DeepSeek V4 Released: Open Source AI Rivals Top Closed Models
3 min readDeepSeek is back — and this time the Chinese AI startup is swinging harder than ever. On April 24, 2026, DeepSeek released preview versions of its V4 model family, positioning it as a direct rival to the best closed-source AI systems on the market.
What Is DeepSeek?
DeepSeek is a Chinese AI research company that made global headlines in early 2025 when it released models that matched GPT-4-class performance at a fraction of the cost. At the time, the release rattled stock markets and forced a rethink of assumptions about AI development costs. The company has been building toward its next major release ever since.
What DeepSeek Just Launched
DeepSeek released two new models: DeepSeek-V4-Pro and DeepSeek-V4-Flash. The Pro variant carries 1.6 trillion parameters with 49 billion activated at any given time using a Mixture-of-Experts architecture. V4-Flash is a lighter option with 284 billion parameters and 13 billion activated.
Both models support a context window of one million tokens, enough to process entire codebases or book-length documents in a single pass. DeepSeek credits a new hybrid attention mechanism for making this possible without ballooning compute costs. Compared to DeepSeek-V3.2, the V4-Pro requires only 27% of single-token inference FLOPs and just 10% of the KV cache at the 1M-token setting.
Pricing is aggressive: V4-Pro costs $3.48 per million output tokens, and V4-Flash drops to just $0.28. Both models will be open source. For the hardware side, DeepSeek partnered with Huawei, whose Ascend 950 chips power V4’s compute clusters using the company’s “Supernode” technology.
Why This Release Matters
The Huawei partnership is the detail that carries the most geopolitical weight. U.S. export controls have restricted DeepSeek’s access to Nvidia chips, yet V4 still claims benchmark performance that rivals OpenAI, Anthropic, and Google. That tells the broader story: American chip restrictions have not stopped China from building frontier AI.
On pure economics, the pricing gap continues to widen. At $3.48 per million tokens for a top-tier model, DeepSeek puts serious pressure on competitors. Open-source availability makes it even harder for Western companies to hold a pricing moat. Developers looking for capable, cost-effective models now have a compelling alternative.
The US-China AI race is no longer a gap story. It is a genuine two-horse competition, and DeepSeek V4 is the clearest evidence yet.
What Comes Next
The current release is a preview, meaning full weights and documentation are still rolling out. Benchmark results and independent third-party evaluations will follow in the coming days. Those results will determine whether V4-Pro’s claims of rivaling closed-source models hold up in real-world tests.
For now, the announcement alone is enough to keep the industry watching carefully.