DeepSeek Just Made Its Best AI Model 75% Cheaper — Forever

On May 23, 2026, DeepSeek quietly posted an update to its API documentation that the AI industry noticed immediately: the 75% discount on its flagship V4-Pro reasoning model is no longer a promotional offer. It is now the permanent price.

Bloomberg and Reuters both confirmed the announcement. What had been framed as a limited-time window since the model's April 2026 launch — a 75% off promotion ending May 31 — will instead become the permanent price structure after that date.

For developers building on AI APIs, this changes the math significantly. For beginners exploring AI tools, it is worth understanding what DeepSeek V4-Pro actually is and whether this price shift means anything for the tools you use every day.

---

What Is DeepSeek V4-Pro?

DeepSeek V4-Pro is the flagship AI model from DeepSeek, a Chinese AI research lab that has repeatedly disrupted the AI industry with high-performance models at unusually low prices. V4-Pro is their most capable model: a reasoning-first system with a 1-million token context window, thinking mode enabled by default, and full support for tool calls and JSON output.

In practical terms, it competes directly with OpenAI's GPT-5.4 and Anthropic's Claude Sonnet 4.6 — the models that most professionals use for serious AI tasks. DeepSeek V4-Pro consistently scores near the top of independent reasoning and coding benchmarks alongside those models.

Until April 2026, accessing models at this capability level meant paying GPT-level prices. DeepSeek changed that.

---

The Price Cut: What It Actually Means

When DeepSeek launched V4-Pro in late April 2026, they immediately offered it at 75% off the listed price as a promotional discount. The official "full price" listed was:

• Input tokens (cache miss): $1.74 per million
• Output tokens: $3.48 per million

The promotional price was a quarter of that. The announcement on May 23 confirmed: after May 31, the pricing will be "officially adjusted to 1/4 of the original price" — meaning the discount becomes the permanent rate:

• Input tokens (cache miss): $0.435 per million
• Output tokens: $0.87 per million
• Input tokens (cache hit): $0.003625 per million

To put those numbers in context:

GPT-5.4 (OpenAI): $2.50 per million input tokens, $10 per million output tokens Claude Sonnet 4.6 (Anthropic): $3 per million input tokens, $15 per million output tokens DeepSeek V4-Pro (DeepSeek): $0.435 per million input tokens, $0.87 per million output tokens

DeepSeek V4-Pro now costs roughly 6x less than GPT-5.4 and roughly 17x less than Claude Sonnet for output tokens, at comparable capability levels for many tasks.

---

Why DeepSeek Can Offer This Price

The price gap is not a marketing gimmick — it reflects genuine structural differences in how Chinese AI labs are built and how they approach distribution.

DeepSeek's training infrastructure is leaner and more efficient than Western counterparts. Their earlier model DeepSeek-R1, released in early 2025, trained for a fraction of the cost of comparable US models by using a mixture-of-experts architecture more aggressively. V4-Pro continues that approach.

There is also a strategic angle. Chinese AI labs are competing for developer mindshare globally, and low API prices are the fastest way to get developers building on your infrastructure. DeepSeek gets adoption data, fine-tuning signal, and ecosystem presence every time a developer calls their API — and at these prices, many will try it.

---

How This Affects What You Should Use as a Beginner

If you are just starting to explore AI tools, the most important thing to understand is the difference between consumer products (ChatGPT, Claude.ai) and developer APIs (OpenAI API, DeepSeek API).

Consumer products like ChatGPT are subscription-based ($20/month) and require no technical setup. You open a browser, type a message, and get a response. The DeepSeek API price cut has no direct effect on these products — ChatGPT pricing is independent of API rates.

Where the DeepSeek price cut matters is if you are:

1. Building something — using AI to power an app, automate a workflow, or process data at scale
2. Using AI-native tools — some productivity apps and developer tools use AI APIs under the hood; cheaper APIs can mean lower product prices over time
3. Exploring DeepSeek directly — DeepSeek's consumer chat interface at chat.deepseek.com is free, similar to ChatGPT's free tier

For complete beginners who want to use AI without writing a single line of code, the pricing change does not mean you need to switch tools. ChatGPT, Claude, and Gemini remain the most accessible consumer interfaces. DeepSeek's chat interface is a solid free alternative worth trying, but the $0.435/M pricing is relevant only when you are calling APIs programmatically.

---

What Happens if You Want to Build With AI

If you are interested in building AI-powered tools for your business or side project, the DeepSeek price cut is genuinely significant.

For developers calling APIs directly, V4-Pro at $0.435/M input tokens is one of the most cost-efficient options for tasks that require genuine reasoning — document analysis, coding, structured data extraction, complex Q&A systems.

For non-developers who want to build custom AI tools without writing code, platforms like CustomGPT let you create AI chatbots and agents for your business using your own documents and data, without touching an API. You define the behavior, they handle the infrastructure. The underlying model costs are absorbed into the platform pricing, not passed to you directly.

---

DeepSeek V4-Pro vs DeepSeek V4-Flash

One nuance worth mentioning: DeepSeek also offers V4-Flash, a faster and cheaper sibling model:

• V4-Flash input: $0.14 per million tokens
• V4-Flash output: $0.28 per million tokens
• V4-Flash context: 1M tokens

V4-Flash is the model behind the deepseek-chat API endpoint (which is being deprecated in favor of deepseek-v4-flash as the explicit name). It is the right choice for high-volume tasks where speed and cost matter more than deep reasoning.

V4-Pro, with thinking mode on by default, is better for tasks that benefit from step-by-step reasoning — math, code review, complex analysis. The permanent 75% price cut makes it accessible for workloads where previously you might have dropped down to V4-Flash to manage costs.

---

The Broader Pattern

This is the third time in 18 months that DeepSeek has moved prices dramatically downward — each time triggering competitive responses from OpenAI, Anthropic, and Google.

When DeepSeek R1 launched in early 2025 at a fraction of GPT-4 prices, it forced a broader reckoning about AI API pricing that has continued through 2026. V4-Pro's permanent price cut is the latest move in that pattern.

For developers and builders, the direct effect is positive: serious reasoning capability now costs less than a dollar per million output tokens. For consumer AI users, the indirect effect is competitive pressure that keeps all the major players from raising prices significantly.

The AI pricing war is not over. But for anyone building on AI today, DeepSeek just locked in a floor that is hard to compete with at the reasoning model tier.

---

FAQ

Q: Does the DeepSeek price cut affect ChatGPT pricing? A: No, they are different products. ChatGPT is a consumer subscription product. DeepSeek V4-Pro is a developer API. OpenAI may respond competitively over time, but there is no direct link between these pricing structures.

Q: Is DeepSeek V4-Pro safe to use for business data? A: This is a legitimate consideration. DeepSeek is a Chinese company and its data handling is subject to Chinese regulations. For sensitive business data, many developers prefer to use it through Azure or AWS Marketplace deployments where data residency and compliance terms are clearer. For general, non-sensitive development work, the API is widely used.

Q: How does DeepSeek V4-Pro compare to Claude Sonnet for coding? A: Both perform well. Claude Sonnet 4.6 has an edge on nuanced instruction following and complex, multi-step reasoning. DeepSeek V4-Pro is competitive on coding benchmarks and significantly cheaper. For most coding tasks, the quality difference is small enough that the price difference dominates the decision.

Q: Can I use DeepSeek V4-Pro through a no-code tool? A: Not directly through most consumer tools yet. The API is available to developers at api.deepseek.com. If you want to build AI tools without code, platforms like CustomGPT let you create custom AI assistants without touching APIs — they handle model selection and infrastructure.

Q: Will DeepSeek cut prices further? A: Historically, yes — DeepSeek has consistently driven prices lower with each generation. V4-Flash already sits at $0.14/M input tokens, which represents where commodity inference is heading. The trend across the industry is toward lower prices as infrastructure costs fall and competition increases.