Alibaba and DeepSeek ignite AI price war with ultra-low token rates, pressuring US rivals

Alibaba on Monday released its flagship Qwen3.8-Max model via API at $2 per 1 million input tokens and $6 per 1 million output tokens, directly undercutting rival Kimi K3 from startup Moonshot AI. DeepSeek has priced its V4-Flash model even lower.

According to benchmark analysis firm Artificial Analysis, DeepSeek V4-Flash costs $0.14 per 1 million input tokens and $0.28 per 1 million output tokens. That equates to about $0.03 per test on standard evaluations, compared with $0.86 for Kimi K3 and $1.86 for OpenAI's GPT-5.6 Sol.

Alibaba talks performance, DeepSeek talks cost

Alibaba claims Qwen3.8-Max, built with 2.4 trillion parameters in a mixture-of-experts architecture, can handle a 1 million token context window. The company says its model tops Moonshot's Kimi K3 on several benchmarks and plans to release the model weights for Qwen3.8-Max and the smaller Qwen3.8-27B next week.

The pricing moves create a tiered market for foundation model APIs:

• Alibaba Qwen3.8-Max: $2 per 1M input tokens
• Moonshot Kimi K3: $3 per 1M input tokens
• DeepSeek V4-Flash: $0.14 per 1M input tokens
• OpenAI GPT-5.6 Sol: approximately $3 per 1M input tokens (inferred per-test cost)

Silicon Valley debates China's growing AI influence

The aggressive pricing from Chinese AI labs has split opinion in Silicon Valley. Some executives view the low-cost open-weight models as a positive force for commoditizing foundation model infrastructure. Others, along with Washington policymakers, frame the trend as a national security challenge to U.S. AI competitiveness.

Alibaba's stock rose 4.5 percent on the day of the Qwen3.8-Max launch, signaling investor confidence that the company can compete effectively. Yet DeepSeek's ultralow pricing suggests that the race to the bottom on inference cost is far from over. Alibaba has committed to open-sourcing the weights of Qwen3.8-Max and its 27B variant next week, which could pressure other labs to follow suit or differentiate on downstream services. The question for enterprise buyers is whether the cheapest model will be good enough for production workloads, or whether reliability and support will command a premium.