DeepSeek V4: Open-Source AI Model Challenges US Dominance with Huawei Chip Support

Chinese AI firm DeepSeek released V4, its flagship open-source model, on Friday, April 24, 2026. The model processes up to 1 million tokens, matching Gemini and Claude, and uses a novel attention mechanism that cuts computing power to 27% and memory to 10% of its predecessor V3.2. V4 comes in two versions: V4-Pro ($1.74 per million input tokens) and V4-Flash ($0.14 per million input tokens), both with reasoning modes. On benchmarks, V4-Pro competes with Anthropic's Claude-Opus-4.6, OpenAI's GPT-5.4, and Google's Gemini-3.1, and exceeds Alibaba's Qwen-3.5 and Z.ai's GLM-5.1 on coding, math, and STEM tasks. Over 90% of 85 surveyed developers ranked V4-Pro among top choices for coding. V4 is DeepSeek's first model optimized for domestic Chinese chips, specifically Huawei's Ascend 950 series, marking a step away from Nvidia. Huawei announced Ascend supernode support for V4. Chinese government officials recommended Huawei integration, aligning with national self-reliance goals amid US export controls since 2022. DeepSeek reportedly gave early access only to Chinese chipmakers, not Nvidia or AMD. However, V4's training still largely relies on Nvidia chips, with Chinese chips used mainly for inference. DeepSeek says V4-Pro prices could fall after Ascend 950 supernodes ship at scale in H2 2026. The release follows months of personnel departures, delays, and scrutiny from US and Chinese governments.