DeepSeek V4-Pro slashes API price 75% with Huawei chip optimization
At $0.83 per million output tokens, this 1.6T-parameter model undercuts rivals.
Get AI news that actually matters
One email a day. Zero fluff. Join 10,000+ professionals.
DeepSeek has permanently slashed the price of its flagship V4-Pro model API by 75%, bringing the cost down to roughly $0.83 per million output tokens. This 1.6-trillion-parameter model delivers SWE-bench scores that rival top closed-source models like those from Google and Anthropic, all while offering a 1M token context window. The price cut is not a short-term promotion but a structural shift driven by a complete overhaul of the underlying hardware infrastructure.
Due to US export controls on advanced Nvidia GPUs, DeepSeek migrated its entire stack to Huawei's Ascend 950 supernodes. By bypassing the restricted, overpriced Nvidia hardware monopoly, they dramatically lowered compute costs and passed the savings directly to developers. This move exposes what many are calling the 'Nvidia tax' — the premium that US AI companies like Google and Anthropic charge to subsidize their dependence on expensive Nvidia chips.
For developers, the implications are immediate. DeepSeek V4-Pro integrates seamlessly with custom setups like Opencode, offers competitive performance at a fraction of the cost, and challenges the assumption that US infrastructure dominance is unassailable. As developers begin migrating heavy reasoning workloads to V4-Pro, the question looms: are the days of paying a 'brand tax' for US-hosted AI over?
- DeepSeek V4-Pro price permanently cut 75% to ~$0.83 per million output tokens.
- Optimized for Huawei Ascend 950 supernodes instead of Nvidia GPUs due to export controls.
- 1.6T parameters, 1M context window, and SWE-bench scores matching top closed-source models.
Why It Matters
This shift challenges US AI dominance by proving cost-effective alternatives exist outside Nvidia hardware.