DeepSeek-V4-Pro Limited-Time 75% Discount: China’s Top Model Kicks Off a Cost War

Yesterday, DeepSeek suddenly announced a limited-time price drop for its V4-Pro model API to just 25% of the original price, effective until May 5. The news exploded across the developer community—many users topped up right away to try it out. Some integrated it with Claude Code in "max" mode and exclaimed, "Now it really feels like a big model!"

This isn’t the first time a Chinese model has triggered a price war, but the timing and scale of this discount are worth paying attention to. V4-Pro is DeepSeek’s flagship model, designed to compete with GPT-4 in performance. Now, with the price cut to a quarter of the original, it further widens the cost-performance gap in favor of Chinese models.

How Big Is the Price Drop?

According to DeepSeek’s official documentation, the original price of the V4-Pro model was ¥14 per million input tokens and ¥28 per million output tokens. With the 75% discount, the costs are now ¥3.5 for input and ¥7 for output. What does this mean in context?

Let’s compare it to major models on the market:

• GPT-4 Turbo: Input $10 per million tokens (≈ ¥72), output $30 (≈ ¥216)
• Claude 3.5 Sonnet: Input $3 (≈ ¥21.6), output $15 (≈ ¥108)
• Gemini 1.5 Pro: Input $1.25 (≈ ¥9), output $5 (≈ ¥36)
• DeepSeek-V4-Pro (after discount): Input ¥3.5, output ¥7

Even without factoring in exchange rates, DeepSeek’s discounted price is more than twice as cheap as Gemini’s, and an order of magnitude cheaper than GPT-4 and Claude. And that’s while V4-Pro is already performing close to GPT-4 levels.

What Developers Are Saying

Community feedback suggests this price drop directly hit the developers' pain points.

Some said they previously found V4-Pro too expensive to use, but now, after topping up ¥100 to test, they found it “very reliable—it smoothly handles my needs,” and that the interaction experience was even better than GPT’s. This sentiment isn’t unique—multiple developers reported that V4-Pro exceeded expectations in code generation and logical reasoning scenarios.

Others made price comparisons and concluded that “Chinese models really max out the cost-performance ratio.” That “maxed out” doesn’t just mean cheap—it means the cost has been lowered to a level where developers can confidently and freely use it without worry, all while maintaining performance.

There’s an interesting detail here: many developers said they use V4-Pro as a “transition.” That shows that GPT-4 and Claude are still considered benchmarks in people’s minds, but in practical use, domestic models can already handle most needs. And with such a clear pricing advantage, “transition” might eventually become “permanent.”

The Logic Behind the Price War

China’s large-model price war began in the latter half of last year. Major players like ByteDance, Alibaba, and Baidu successively lowered prices—some models even dropped to below ¥1 per million tokens. On the surface, DeepSeek’s latest price cut looks like it’s following the trend, but there’s strategic thinking behind it.

First, since V4-Pro is DeepSeek’s flagship model, lowering the price helps rapidly expand its user base. For a relatively young model, getting more developers to use it and build reputation is more important than short-term revenue.

Second, DeepSeek has worked extensively on inference efficiency optimizations. Since last September, they’ve been gradually lowering API prices. Reaching a 75% discount now suggests strong confidence in cost control. Architectural optimizations and inference acceleration technologies have given them room to wage a price war.

At a deeper level, this price war reflects the competition for market share among domestic models. International ones like GPT-4 and Claude face restrictions in China, so domestic models have a natural localization advantage. But to really get developers to switch, “usable” isn’t enough—it has to be “good” and “cheap.” DeepSeek’s price cut is a strategic strike under this logic.

Beyond the Cost-Performance Ratio

Of course, price isn’t everything.

Technically, V4-Pro performs well in code generation, long-text understanding, and multi-turn conversations. However, in specialized professional domains such as medicine and law, it may still lag behind GPT-4 or Claude. This isn’t unique to DeepSeek—it’s a general challenge for domestic models, which still need time to build sufficiently broad and deep training data coverage.

API stability is another concern. Domestic models often lag behind established players like OpenAI and Anthropic in high-concurrency scenarios. If your application depends heavily on response time and availability, you may need to weigh those factors carefully.

Another often-overlooked factor is model update frequency. OpenAI and Anthropic release improvements every few months, continuously refining performance. Domestic models operate at a slower pace—V4-Pro hasn’t had a major update since release. If your project depends on staying at the forefront of new technology, this is worth considering too.

What It Means for Developers

This price cut is great news for developers, especially individuals and small teams sensitive to costs.

If you're building experimental projects, internal tools, or anything not requiring top-of-the-line performance, V4-Pro now offers outstanding value. At ¥3.5 per million input tokens, you can run extensive tests and product iterations for very low cost.

For production environments, it’s best to test first. Route some traffic to V4-Pro to evaluate its actual performance and stability before deciding on a full switch. Since the discount is time-limited—ending May 5—you’ll want to reassess later whether the regular price still fits your budget.

If you’re using an API aggregation platform (like OpenAI Hub), you can easily switch to DeepSeek-V4-Pro to try it out. Such platforms usually support multi-model switching, so you can compare different models’ results without changing your code.

How Long Will the Price War Last?

That’s hard to say, but there are a few trends to watch.

First, the domestic price war will continue. All the major providers are competing for market share, and price cuts remain the most direct tool. Still, there’s a bottom line—model training and inference come with real costs. Sustained losses aren’t feasible, so future pricing may favor limited-time promotions instead of permanent cuts.

Second, the balance between performance and price will be key. Competing on price alone is meaningless—capability ultimately matters most. Companies like DeepSeek, Zhipu, and Kimi are all continuously improving their models, and we may see “tiered pricing by performance” emerge—cheaper base editions and pricier premium editions, allowing users to choose based on their needs.

Finally, international models may also adjust prices. If domestic competitors keep gaining market share, OpenAI and Anthropic may feel the pressure. While they’re unlikely to slash prices directly (given their brand premium), they might introduce region-specific discounts or bundle deals.

Final Thoughts

On the surface, DeepSeek-V4-Pro’s price cut is a promotional campaign, but in essence, it signals a new stage in the competition among Chinese AI model providers. The price war benefits developers but will also hasten industry consolidation—companies without technical depth and relying solely on spending will be eliminated, while truly capable players will stand out.

For developers, this is a great time to experiment. Take advantage of the discount period to test, compare, and find the best-fitting model for your projects. Don’t focus solely on price—performance, stability, and ecosystem support are all crucial.

The rise of Chinese models is an encouraging sign, but optimism should be tempered with realism. Building technical depth takes time; earning market trust takes reputation. Whether DeepSeek’s price cut can truly shift the market will become clear in the coming months.

---

Sources

• DeepSeek V4-Pro Official Discount Announcement – Limited-time discount details in the official DeepSeek API documentation
• DeepSeek V4-Pro Cost-Performance Discussion – Developer community analysis of the discount’s cost-effectiveness
• DeepSeek V4-Pro User Experience Sharing – Developer feedback and reviews after hands-on use