Sign up to save your podcasts
Or
Share DeepSeek V4 Preview Launch Challenges US AI Dominance with Cost Effective Models and Huawei Chip Integration
Share to email
Share to Facebook
Share to X
Share to email
Share to Facebook
Share to X
Or
DeepSeek V4 Preview Launch Challenges US AI Dominance with Cost Effective Models and Huawei Chip Integration
Listeners, China's DeepSeek has just launched preview versions of its highly anticipated V4 AI models, marking a bold step in the intensifying US-China AI rivalry. According to 1News, the rollout happened Saturday, introducing V4 Pro and V4 Flash open-source versions with massive upgrades in knowledge, reasoning, and agentic capabilities—the ability to handle complex tasks autonomously.
DeepSeek's official API docs announce that DeepSeek-V4 Preview is now live and open-sourced, boasting a cost-effective 1 million token context window, a huge leap from V3's 128,000 tokens. The V4 Pro packs 1.6 trillion total parameters but activates only 49 billion per token for frontier-level performance at lower compute costs, while the Flash version, with 284 billion total and 13 billion active, offers speed and efficiency.
The Jakarta Post reports close collaboration with Huawei, whose Ascend chips powered parts of V4's training, cutting reliance on US Nvidia chips—a shift from DeepSeek's past models. DW News highlights this as a challenge to OpenAI and Anthropic, with V4 Pro Max claiming superior reasoning over OpenAI's GPT-5.2 and Google's Gemini 3.0-Pro, though marginally behind GPT-5.4 and Gemini 3.1-Pro. In agentic tasks, V4 Pro outperforms Anthropic's Claude Sonnet 4.5 and nears Opus 4.5.
The National News notes V4 Pro runs about 50 times cheaper than Claude Opus via API, potentially pressuring Nvidia's dominance, especially as it launched when Nvidia hit a $5 trillion cap. Japan Times points out it doesn't fully close the US lead but underscores China's progress, aligning with the 2026 Stanford AI Index calling the performance gap effectively closed.
These previews allow real-world feedback before final release, with no timeline yet. DeepSeek's free chatbot now defaults to 1M context across services.
Thanks for tuning in, listeners—come back next week for more. This has been a Quiet Please production, and for more, check out Quiet Please Dot A I.
Some great Deals https://amzn.to/49SJ3Qs
For more check out http://www.quietplease.ai
This content was created in partnership and with the help of Artificial Intelligence AI.
View all episodes
By Inception Point AIListeners, China's DeepSeek has just launched preview versions of its highly anticipated V4 AI models, marking a bold step in the intensifying US-China AI rivalry. According to 1News, the rollout happened Saturday, introducing V4 Pro and V4 Flash open-source versions with massive upgrades in knowledge, reasoning, and agentic capabilities—the ability to handle complex tasks autonomously.
DeepSeek's official API docs announce that DeepSeek-V4 Preview is now live and open-sourced, boasting a cost-effective 1 million token context window, a huge leap from V3's 128,000 tokens. The V4 Pro packs 1.6 trillion total parameters but activates only 49 billion per token for frontier-level performance at lower compute costs, while the Flash version, with 284 billion total and 13 billion active, offers speed and efficiency.
The Jakarta Post reports close collaboration with Huawei, whose Ascend chips powered parts of V4's training, cutting reliance on US Nvidia chips—a shift from DeepSeek's past models. DW News highlights this as a challenge to OpenAI and Anthropic, with V4 Pro Max claiming superior reasoning over OpenAI's GPT-5.2 and Google's Gemini 3.0-Pro, though marginally behind GPT-5.4 and Gemini 3.1-Pro. In agentic tasks, V4 Pro outperforms Anthropic's Claude Sonnet 4.5 and nears Opus 4.5.
The National News notes V4 Pro runs about 50 times cheaper than Claude Opus via API, potentially pressuring Nvidia's dominance, especially as it launched when Nvidia hit a $5 trillion cap. Japan Times points out it doesn't fully close the US lead but underscores China's progress, aligning with the 2026 Stanford AI Index calling the performance gap effectively closed.
These previews allow real-world feedback before final release, with no timeline yet. DeepSeek's free chatbot now defaults to 1M context across services.
Thanks for tuning in, listeners—come back next week for more. This has been a Quiet Please production, and for more, check out Quiet Please Dot A I.
Some great Deals https://amzn.to/49SJ3Qs
For more check out http://www.quietplease.ai
This content was created in partnership and with the help of Artificial Intelligence AI.