DeepSeek Redraws the Coordinate System with V4

An Asymmetric Pricing Manifesto

In the summer of 2025, the AI industry witnessed a scene of extraordinary drama: OpenAI announced API price increases, pushing the cost of its high-end models even higher. Almost simultaneously, DeepSeek released its V4 model and permanently slashed its cache hit price to one-tenth of the original.

Two companies. Two directions. The price gap for comparable models stretched to 34.5x.

That number alone is staggering, but to interpret this merely as a "price war" would be to underestimate the true significance of the shift.

Behind the 34.5x Gap: Two Fundamentally Different Paths

Let's start with the numbers. What OpenAI has been consistently doing is raising the ceiling on model capability while maintaining — or even increasing — its pricing power. The logic is clear: the stronger the model, the deeper the moat, the more dependent the users, and the more sustainable the pricing. This is the classic "platform" playbook — pull developers into the ecosystem, lock workflows to APIs, and build lock-in through brand and inertia.

DeepSeek has taken a different road. V4 doesn't just match or exceed comparable competitors in performance; more critically, it declared its stance with a pricing strategy that borders on the "irrational" — cutting cache hit prices to 1/10, meaning costs in high-frequency calling scenarios become nearly negligible. This isn't a promotional discount. It's a structural elimination of the "API tax."

One is building walls higher. The other is tearing them down. The directions are diametrically opposed.

Not a Price War — a Divergence in Philosophy

To understand this shift, we need to step beyond the "who's cheaper" framework.

OpenAI's price-hike logic is understandable. The company has invested tens of billions of dollars in training costs, carries investor expectations for returns, and is building a full-stack platform from model to application. ChatGPT's subscription model, tiered API pricing, deep enterprise integration — every move reinforces the "platform" identity. The deeper users go, the higher the migration costs. This is the core flywheel of platform economics.

DeepSeek's V4 presents an "anti-platform" posture. It doesn't build a closed ecosystem, doesn't design lock-in mechanisms, and even proactively pushes prices so low that "lock-in" becomes meaningless. When calling costs drop below a certain threshold, developers no longer need to stay on a particular platform to save money. When the model itself is open-source or semi-open-source, users no longer need to surrender data sovereignty to access capabilities.

This divergence ultimately answers the same fundamental question: Where should AI's value accumulate?

OpenAI's answer: In the platform. The model is the gateway; the ecosystem is the moat.

DeepSeek's answer: At the application layer. The model is infrastructure — it should be as cheap and ubiquitous as water and electricity.

The Map Is Being Redrawn

This divergence is reshaping the competitive map of the entire industry.

For developers, the 34.5x price gap isn't just a cost calculation — it's an architectural decision. With DeepSeek V4 pushing inference costs to the extreme, many application scenarios that previously "didn't pencil out" suddenly become viable: intelligent customer service with multi-turn dialogue, real-time code review, large-scale document processing, hybrid edge-cloud inference. These scenarios are extremely price-sensitive. A cost reduction of one order of magnitude means a market expansion of one order of magnitude.

For other model providers, the imperative is to pick a side — fast. Follow OpenAI's platform route, justifying premium pricing through differentiated services and deep integration? Or follow DeepSeek's infrastructure route, competing on scale through extreme efficiency and low prices? The middle ground is vanishing.

For the AI industry as a whole, this may be a turning-point signal. Over the past two years, the dominant narrative has been the "model capability race" — whoever has the most parameters, whoever scores highest on benchmarks, wins. But DeepSeek V4 demonstrates through action that as model capabilities converge, the axis of competition shifts: from "who's more powerful" to "who's cheaper," and then to "who can make it accessible to the most people."

The Real Question

Back to the most fundamental question: How can DeepSeek afford this price?

The answer isn't simply "willingness to earn less." V4's architectural optimizations, inference efficiency improvements, and caching mechanism refinements have driven its marginal costs genuinely far below those of its peers. In other words, this isn't a subsidized low price — it's an engineering-driven low price. And that's what competitors should find most unsettling.

Subsidies can be exhausted, but efficiency advantages are self-reinforcing — more users mean higher cache hit rates, which mean lower costs, which enable lower prices, which attract more users. This is a positive feedback loop entirely different from the platform flywheel, but equally powerful.

Outlook: The Coordinate System Has Changed

What DeepSeek V4 has done is not find a lower price point within the old coordinate system — it has redefined the axes entirely. While the industry was still mapping competition along "capability × price," DeepSeek shifted the coordinate system to "efficiency × openness."

In this new coordinate system, OpenAI remains formidable, but its position has changed. It is no longer the sole reference origin, but rather one route among many — perhaps the most premium one, but not necessarily the one most will travel.

The map is being redrawn. And this time, the pen isn't in Silicon Valley's hand.