DeepSeek has released V4, a new open-source AI model that dramatically undercuts Western competitors on price while delivering performance comparable to Anthropic's Claude Sonnet and approaching Claude Opus in some coding tasks. The Chinese AI startup unveiled the preview on April 24, introducing a model that costs roughly 10 times less than OpenAI's GPT-5.5 for the same computational work. The release marks a significant shift in the global AI landscape, particularly because V4 is optimized to run on Huawei's domestic chips rather than Nvidia hardware.

How Does DeepSeek V4 Achieve Such Lower Costs?

DeepSeek's cost advantage stems from a technical innovation called a hybrid attention mechanism, which combines Compressed Sparse Attention and Heavy Compressed Attention. This approach dramatically reduces the computing power and memory required to process long documents and conversations. At 1 million tokens, V4-Pro requires only 27 percent of the inference computing operations and just 10 percent of the key-value cache compared to DeepSeek's previous V3.2 model.

To put this in practical terms, a 10-fold reduction in memory cache means a single graphics processing unit (GPU) can handle roughly 10 times as many simultaneous long-context sessions. DeepSeek has made 1-million-token context the default for its API rather than a premium tier, meaning users can process roughly 100,000 words at once without paying extra.

• V4-Pro Version: Contains 1.6 trillion total parameters with 49 billion active parameters, designed for more demanding tasks and reasoning-heavy workflows
• V4-Flash Version: Contains 284 billion total parameters with 13 billion active parameters, optimized for speed and efficiency on resource-constrained systems
• Availability: Both versions are available now on Hugging Face and through DeepSeek's API, with the final version still pending announcement

How Does V4 Compare to Claude and Other Leading Models?

DeepSeek claims V4-Pro outperforms all other open-source models on world knowledge benchmarks, trailing only Google's Gemini 3.1 Pro. In agentic coding evaluations, which test a model's ability to write and debug code autonomously, V4-Pro beats Claude Sonnet 4.5 and approaches Claude Opus 4.6's non-thinking mode performance.

On competition mathematics benchmarks, V4-Pro-Max, the version with maximum reasoning effort, scored 95.2 on the HMMT 2026 February exam and 89.8 on IMOAnswerBench, placing it within range of GPT-5.4's scores of 97.7 and 91.4 respectively. However, DeepSeek acknowledges that V4 "falls marginally short of GPT-5.4 and Gemini 3.1 Pro, suggesting a developmental trajectory that trails state-of-the-art frontier models by approximately three to six months".

"DeepSeek-V4 Preview is officially live and open-sourced. Welcome to the era of cost-effective 1M context length," the company announced.

DeepSeek, Official Announcement

What Makes the Huawei Chip Integration Significant?

The most consequential aspect of V4's release is its optimization for Huawei's Ascend AI chips rather than Nvidia hardware. DeepSeek trained its earlier V3 and R1 models on Nvidia H800 GPUs, but V4's pivot to domestic Chinese chips comes as U.S. export controls continue blocking Chinese developers from purchasing Nvidia's most advanced processors.

Hours after the preview launched, Huawei confirmed that V4 is fully supported on its Ascend 950-based supernode clusters and that its chips were used for part of V4-Flash's training. The company stated that "the entire Ascend supernode product line now supports the DeepSeek-V4 series models".

Huawei

"This is a big deal for China's AI industry. Huawei's Ascend chips are the country's best homegrown alternative to Nvidia, and supporting DeepSeek V4 shows that top Chinese AI models can now run on Chinese hardware," explained He Hui, director of semiconductor research at Omdia.

He Hui, Director of Semiconductor Research at Omdia

DeepSeek expects V4-Pro pricing to drop sharply once Huawei scales up Ascend 950 production in the second half of 2026. Currently, V4-Pro costs up to 12 times more than V4-Flash due to constraints in high-end compute capacity, but this gap should narrow significantly as manufacturing capacity increases.

What Are the Practical Implications for Developers?

V4 is designed for agent workflows, not just chatbot interactions. DeepSeek says the model works with mainstream development frameworks, including Claude Code, OpenClaw, OpenCode, and CodeBuddy, allowing developers to integrate it into existing coding tools. The company has been using V4 internally as its primary agentic coding model for program development tasks, optimizing it for multi-step processes like data collection, organization, and output generation as a complete workflow rather than isolated responses.

Developers should note that DeepSeek's existing deepseek-chat and deepseek-reasoner API endpoints will be fully retired after July 24, 2026. Developers need to migrate to the explicit deepseek-v4-pro and deepseek-v4-flash model IDs before that date to avoid service interruptions.

What Does This Mean for the Broader AI Industry?

V4's release arrives amid escalating geopolitical tensions around AI development. The same day the preview launched, Reuters reported that the U.S. State Department sent a diplomatic cable instructing embassy staff worldwide to warn foreign governments about alleged intellectual property theft by DeepSeek and other Chinese AI firms. The White House Office of Science and Technology Policy published a memo earlier in the week accusing Chinese entities of running "deliberate, industrial-scale campaigns" to distill American frontier AI systems.

Reuters

Anthropic has claimed that DeepSeek, Moonshot AI, and MiniMax used 24,000 fraudulent accounts to make 16 million exchanges with its Claude model. China's foreign ministry called the accusations "groundless" and "a smear against the achievements of China's AI industry".

Beyond the geopolitical dimension, V4's pricing and performance suggest that the cost of deploying advanced AI models is dropping faster than many industry observers anticipated. DeepSeek is reportedly raising funds at a valuation exceeding 20 billion dollars, with Alibaba and Tencent in discussions to take stakes, signaling strong investor confidence in the company's trajectory.

" }