May 6, 2026
Discount or plot twist?
DeepSeek V4 Pro at 75% off until 31 May
AI prices were supposed to go up — so why is DeepSeek suddenly going bargain-bin cheap
TLDR: DeepSeek slashed the price of its V4 Pro AI tool by 75% for a limited time, making an already cheap service even cheaper. The comments instantly split between bargain-hunters cheering the deal, skeptics questioning safety, and working developers asking whether the low-cost option is actually good enough to use every day.
DeepSeek tried to make a simple pricing update, but the crowd turned it into a full-on "wait, I thought the cheap AI party was over?" moment. The company says its DeepSeek V4 Pro model is now 75% off for a limited time, with even lower prices when the system can reuse earlier text instead of reading everything again. In plain English: people who use these tools to write, code, or answer questions may be able to do a lot more for a lot less — at least until the promo ends. And yes, users immediately started doing the math in public like they were clipping coupons at a supermarket.
The loudest reaction was disbelief. One commenter basically summed up the mood: just days after people were declaring that the era of dirt-cheap AI was dead, here comes a giant markdown and bigger usage limits. Another user posted the price breakdown and declared it "damn cheap," which is pretty much the thread’s unofficial slogan.
But not everyone was in a celebratory mood. One of the spiciest comments raised concerns about China’s National Intelligence Law, shifting the vibe from bargain hunting to geopolitical side-eye. That sparked the classic internet split: some people are staring at the discount, others are staring at the risk.
Then came the practical crowd. One developer said DeepSeek V4 has been very very good for real work, even on a weirdly impressive old-school game reverse-engineering project, while another asked the painfully relatable question: great price, but what coding helper actually works well with it? In other words, the community verdict is messy, dramatic, and extremely online: cheap enough to tempt everyone, controversial enough to keep the comments cooking.
Key Points
- •DeepSeek prices its models per 1 million tokens, with billing based on total input and output tokens.
- •The model names deepseek-chat and deepseek-reasoner are planned for future deprecation.
- •For compatibility, deepseek-chat and deepseek-reasoner currently correspond to non-thinking and thinking modes of deepseek-v4-flash.
- •DeepSeek says input cache hit pricing for all models has been reduced to one-tenth of the launch price.
- •DeepSeek is offering a 75% discount on deepseek-v4-pro until 2026/05/05 15:59 UTC, and fees are deducted from granted balance first when available.