China is witnessing a fierce pricing and parameter war among its top artificial intelligence companies. DeepSeek is preparing to launch its V4 model tomorrow to challenge heavyweights like Alibaba and Moonshot. But the biggest talking point is a wild new pricing strategy that charges users more during peak traffic hours.
The upcoming release is divided into 2 versions. The premium option is V4 Pro, while the smaller one is V4 Flash. Early testers report that this architecture delivers massive upgrades in programming tasks and agent creation. In terms of raw capabilities, the V4 Pro version is currently matching premium international systems like Opus 4.8. Its coding proficiency even rivals GPT 5.6 Sol from OpenAI. Some testers think V4 might fall slightly short of the newly released Kimi K3 in absolute power, but it easily wins on pricing.
According to reports circulating across domestic media, DeepSeek is introducing a peak and valley pricing model for its API. This means the cost of using the system swings based on network traffic. The V4 Pro version charges $0.87 per 1 million output tokens during quiet hours but spikes to $1.74 during busy windows. For the lighter V4 Flash version, users pay just $0.28 per 1 million tokens during off peak times and $0.56 during high traffic hours. Developers will need to schedule their heaviest computing tasks carefully to keep costs down.
Alibaba has opened public previews for its Qwen 3.8 Max model to keep pace with the competition. This system boasts 2.4 trillion parameters, focusing on deep reasoning and programming engineering. It represents a broader trend where domestic developers are racing past the trillion parameter mark. Kimi K3 leads the pack at 2.8 trillion parameters, while MiniMax is working on its M3 Pro model at 2.7 trillion. Even with smaller models like the 1.6 trillion parameter V4 Pro, developers are finding success through clever post training optimizations. Parameter count is no longer the only metric that matters.
