DeepSeek and Alibaba launch fresh assaults on frontier AI
Two of China's biggest AI labs spent the weekend taking aim at frontier models.
Image by Cybernews.
- DeepSeek launched V4 and Alibaba unveiled Qwen 3.8, claiming both approach top US AI models.
- Both companies are using lower pricing to challenge OpenAI, Anthropic, and other frontier AI providers.
- Alibaba’s Qwen 3.8 has 2.4 trillion parameters, making it one of the largest open-weight AI models released.
Key Takeaways by nexos.ai, reviewed by Cybernews staff.
DeepSeek released V4 into general availability, while Alibaba unveiled its open-weight Qwen 3.8. Together, the companies claim their latest models rival or approach leading systems from Anthropic and OpenAI, underscoring China's push to compete at the cutting edge of AI.
Early testers report V4's performance approaching that of Anthropic's Claude Opus 4.8 and OpenAI's GPT-5.6 Sol, at a fraction of the cost. Alibaba says the open-weight Qwen 3.8 is "second only to Fable 5."
DeepSeek's V4 went from preview to general availability on July 19th, 3 months after its initial release. Even in preview, its coding score sat at 80.6% on SWE-bench Verified, closing in on Anthropic's Opus 4.6.
Hours after DeepSeek's rollout, the Qwen team announced Qwen 3.8. At 2.4 trillion parameters, it would be one of the largest open-weight models ever released. A preview is already live on Alibaba's Token Plan and its Qoder coding platforms.
Stay updated with our latest stories and follow us on social media
Be the first to discover new stories, ideas, and updates from our team.
A clear direction
DeepSeek introduced dynamic pricing alongside the launch. During off-peak hours, V4 Pro runs 87 cents per million tokens, against $50 for the same volume from Anthropic's Fable 5.
Similarly, the Qwen 3.8 preview is priced at a tenth of Alibaba's standard rates. There's no benchmark available yet, but the intention is clear: match the frontier, then undercut it on cost.
The two weekend announcements follow the release of Kimi K3 by Chinese startup Moonshot. K3 is a 2.8-trillion-parameter open-weight model that the company claims performs close to Fable 5.
DeepSeek's 75% discount on V4 Pro started as a temporary promotion. In May, the company made it permanent, undercutting GPT-5, Gemini, and Claude outright.
The Cybernews community is talking about this. Be a part of the conversation.
Beyond competitive pricing, China is also rebuilding its AI stack around indigenous chips.
When DeepSeek V4 first launched in preview, the company optimized it for Huawei's Ascend line instead of Nvidia's. DeepSeek has reportedly been developing its own inference chip since last year, aiming to cut its hardware dependence on Nvidia and Huawei altogether. Alibaba is running the same playbook with its own Zhenwu M890 AI chip.
While the benchmarks haven't been verified by independent leaderboards, the larger pattern isn't only about scores. Chinese AI labs are closing the gap to frontier AI models, undercutting them on price, and increasingly building on chips that Washington can't restrict.
