DeepSeek has paired its open source V4-Pro model with Huawei hardware while cutting prices sharply, signalling a push to challenge proprietary AI leaders through cost, scale and sovereign infrastructure.
DeepSeek has launched V4-Pro, positioning the 1.6-trillion-parameter system among the largest open-source AI models while tying it to Huawei hardware in a move that strengthens domestic chip ecosystem integration and adds a sovereignty dimension to the open-source challenge.
Built on a Mixture-of-Experts architecture that activates around 49 billion parameters per task, V4-Pro aims to deliver large-model performance with lower compute costs. DeepSeek said the model outperforms most open-source rivals on world-knowledge benchmarks and is optimised for AI agents and complex multi-step reasoning.
To accelerate adoption, the company is offering developers a 75% discount on V4-Pro until May 5 and has cut input cache-hit API pricing to one-tenth of previous levels, positioning itself as a lower-cost alternative to proprietary rivals including OpenAI, Google and Anthropic.
The strategy combines open-source scale with pricing pressure at a time when developers are scrutinising inference costs for agentic and high-volume applications.
The Huawei adaptation broadens the significance of the launch beyond model performance, linking open-source AI development with sovereign hardware ecosystems.
Early developer feedback has also been positive, with many users reportedly considering V4-Pro as a default coding model and noting performance close to leading proprietary systems.
The launch signals that competition in AI may increasingly hinge not only on model capability, but also on open infrastructure and economics.
