Two Price Cuts in Two Days: What Is DeepSeek Really Playing At?
Two Price Cuts in Two Days: What Is DeepSeek Really Playing At? On April 24, DeepSeek released the V4-Pro flagship model and announced a 75% API price
Two Price Cuts in Two Days: What Is DeepSeek Really Playing At? On April 24, DeepSeek released the V4-Pro flagship model and announced a 75% API price cut. Just two days later, on the evening of April 26, it announced again: the cached-input price across the entire lineup dropped to 1/10 of the launch price. Stacked with a limited-time 75% discount, V4-Pro cached input fell as low as ¥0.025 per million tokens — a new global low for large-model pricing. This is no ordinary price war. As an AI company whose valuation has jumped from $10 billion to $20 billion, DeepSeek's successive cuts are backed by a complete business logic — from open-source models to API price cuts, from compute substitution to an application explosion, every step is carefully laid out. Today we won't chase the headlines; we'll break down what this price cut really means and what opportunities brand marketers can read from it. Price Cuts Aren't the Goal — Winning Developers Is Consider the numbers: after the cut, V4-Flash cached input dropped from ¥0.2 to ¥0.02 per million tokens, a 90% drop. V4-Pro, with the 75% discount before May 5, is just ¥0.025. Output pricing also dropped 75%, from ¥24 to ¥6 per million tokens. What does this price mean? It's an order of magnitude cheaper than OpenAI and Anthropic models at the same tier. For high-frequency API use cases — customer-service bots, batch content generation, data analysis — cost drops by double digits. AI projects previously shelved over cost concerns can now restart. Brands can use AI more boldly for A/B testing, multi-channel content adaptation, and user-profiling analysis without worrying about budget overruns. The core logic of the cuts: first prove model capability with V4-Pro, then use price to seize developers and enterprise customers at scale. The V4 series has 1.6 trillion total parameters and expanded its context window from 128k to 1 million tokens — technically capable of complex tasks. The price cut is just the catalyst to get more people using it. "Open source made money; closed source isn't convinced yet" — as a recent 36Kr headline put it. Three Takeaways for Brand Marketers • The cost barrier to AI content production is disappearing. The old concern with batch content generation was API cost; now at a few cents per million tokens, that concern is gone. Brands can use AI more aggressively for A/B testing, multi-channel adaptation, and user profiling. A copy A/B test that once took hundreds of API calls was a real cost constraint; now it's negligible. Marketing teams can experiment more boldly, no longer boxed in by budget. • Model-capability competition has entered deep waters. The V4 series supports 1 million tokens of context (previously 128k) plus 1.6 trillion total parameters, meaning it can handle more complex, more specialized marketing tasks — not just simple copy generation, but understanding brand tone, analyzing competitor strategy, and generating complete marketing plans. When a model can hold tens of thousands of words of brand docs and competitor analysis, output quality is completely different. • A domestic AI ecosystem is taking shape. From open-source models to API price cuts to accelerating domestic-chip substitution, the whole chain is forming a complete system. When brands choose AI tools, domestic options have moved from "backup" to "top choice" — not just for price, but because the ecosystem, toolchain, developer community, and industry cases are all accumulating fast. This isn't piecemeal progress; it's systematic catch-up. When to Use It — and When Not to Chase It Low prices don't mean use it for everything. Brand marketers must judge: if your core need is high-frequency content generation — multi-platform posting, A/B copy testing, data reports — this is the best window to use domestic large models. These scenarios are light on creativity but sensitive to efficiency and cost, exactly where large models shine. But if your brand demands extreme tone control and strong human creative oversight, AI remains just an aid and can't fully replace human judgment. Good marketing was never just about content speed — it's insight into human nature, grasp of trends, and understanding of brand soul. AI can't replace these anytime soon. The right approach: let AI handle efficiency and scale, let humans steer direction and tone. A recent Lyon research note also noted: as Agentic AI and multimodal applications spread, token usage is surging and high-end chip supply stays tight. Low price is one thing; whether compute can keep up is another. The dividend window won't stay open forever — run your workflows through while cost is lowest. Three Things to Do Right Now • Test your existing AI workflows. Swap your current AI tool for DeepSeek V4 and compare quality and speed. Use the 75% discount window (before May 5) to cut costs first. No need for a big-bang switch — pick one non-core scenario for a comparison test, then decide on full migration. • Reassess content-capacity upgrades. If API cost previously capped your output, re-plan now. Consider using AI to expand coverage — more RED posts, new platforms, more localized content in more languages. Cost is no longer the bottleneck; capacity is. When a single call costs a few cents, you should think not about saving money but about investing the savings in more innovation. • Watch the domestic-AI substitution trend. Not just the models, but the supporting toolchain, developer community, and industry cases. Early movers will lead the next round of competition. This isn't a simple price cut — it's a signal that domestic AI is shifting from catching up to leading. For marketing-tech teams, now is the best time to plan next-generation AI workflows. A changing world demands deep insight. See you next time.
