DeepSeek has introduced its V4-Flash model, which is making waves due to its competitive performance and significantly lower pricing compared to established models like Anthropic's Claude Opus and OpenAI's Luna. The V4-Flash-0731, priced at $0.14 per million input tokens and $0.28 per million output tokens, is substantially cheaper than Claude Opus 4.8 ($5.00 per million input, $25.00 per million output) and OpenAI's Luna (listed at $0.20 per million input, $1.20 per million output). While some initial claims about V4-Flash beating its own flagship V4-Pro on all benchmarks were later nuanced, the model demonstrates strong performance in specific areas, such as a Terminal Bench 2.1 score of 82.7, and represents a compelling option for developers prioritizing cost and context length.
Alibaba has also launched its Qwen3.8-Max-Preview, a 2.4-trillion-parameter model that the company claims is second only to Anthropic's Claude Fable 5 in overall performance. This release highlights China's rapid advancements in AI, with Alibaba making the model available on its coding platforms and committing to releasing open weights. The announcement led to a 5.4% climb in Alibaba's shares in Hong Kong, reflecting increased investor confidence in its AI capabilities and positioning it as a serious contender in the global AI race.
The competitive pricing strategy of DeepSeek, especially with its V4-Flash model, aims to disrupt the market by making powerful AI more accessible for high-volume applications like coding assistants and document processing tools, where output pricing can significantly impact monthly bills. For instance, DeepSeek's V4-Flash-0731 offers a blended cost per benchmark point of $0.0020 on Terminal Bench 2.1, dramatically undercutting GLM-5.2 ($0.0247) and Claude Opus 4.8 ($0.1059).
Both DeepSeek and Alibaba's new offerings underscore a growing trend where Chinese AI companies are not only developing large and capable models but also strategically pricing them to gain market share. This intensifies the competition with Western counterparts like Anthropic and OpenAI. Alibaba's rapid development cycle, from its original Qwen models in 2023 to the 2.4-trillion-parameter flagship in 2026, also suggests that U.S. export controls on advanced semiconductors may be less effective than intended in slowing China's AI progress.
