According to the official DeepSeek API documentation, once the promotion ends the model’s API pricing will be permanently adjusted to one‑quarter of its original level, effectively making the discounted pricing the new standard rate.

Before the discount, the listed base price was roughly $1.74 per million input tokens and $3.48 per million output tokens, meaning the current pricing represents a lasting step‑change rather than a short‑term marketing promotion.

How it compares with GPT‑5.5 and Claude Opus 4.7

DeepSeek’s pricing stands out most clearly when compared with other frontier AI models.

OpenAI GPT‑5.5

• Input: $5.00 per million tokens (short context)
• Output: $30.00 per million tokens

Anthropic Claude Opus 4.7

• Input: $5.00 per million tokens
• Output: $25.00 per million tokens

DeepSeek V4‑Pro

• Input: $0.435 per million tokens
• Output: $0.87 per million tokens

That means:

• Input tokens are roughly 91% cheaper than GPT‑5.5 or Claude Opus 4.7.
• Output tokens are roughly 96–97% cheaper than those flagship models.

For developers running high‑volume workloads—such as coding assistants, agents, document analysis pipelines, or customer‑support systems—token pricing often dominates total operating costs. A difference of this magnitude can change which architectures are economically viable.

Large‑context capabilities

Beyond pricing, V4‑Pro also competes technically with frontier models.

Key specifications include:

• 1‑million‑token context window
• Maximum output of up to 384,000 tokens
• Support for structured JSON output and tool/function calling

The model is built using a Mixture‑of‑Experts (MoE) architecture with roughly 1.6 trillion total parameters and about 49 billion active parameters per inference step, enabling high capacity without proportionally higher compute costs.

Large context windows are especially useful for:

• long‑document analysis
• codebase‑scale reasoning
• agent workflows with persistent context
• retrieval‑augmented generation with large knowledge bases

Combined with extremely low token pricing, these capabilities enable use cases that would be prohibitively expensive with higher‑priced models.

Why this matters for developers

For many teams, inference cost—not model capability—is the limiting factor when deploying AI at scale. Ultra‑low token prices allow developers to:

• run larger prompts and longer conversations
• generate longer outputs without cost spikes
• process huge document sets or codebases
• build multi‑agent pipelines with many model calls

In practical terms, the difference between $0.87 and $25 per million output tokens can turn previously experimental workflows into economically viable products.

The broader AI pricing war

DeepSeek’s strategy reflects a larger shift underway in the AI industry.

Historically, frontier models were priced in the range of several dollars to tens of dollars per million tokens, as seen with GPT‑5.5 and Claude Opus 4.7.

DeepSeek instead pursues a volume‑driven pricing model, offering dramatically cheaper inference while maintaining competitive capability. Analysts note that its models have repeatedly undercut incumbent providers by large margins, helping push overall AI costs downward across the market.

This approach mirrors dynamics seen in cloud computing and GPUs: once a vendor proves high‑performance systems can run much cheaper, the entire market is forced to respond.

What to watch next

DeepSeek’s pricing move raises several strategic questions for the AI ecosystem:

• Will competing labs lower API prices to remain competitive?
• Will cheaper inference accelerate the shift toward agent‑based applications?
• Can ultra‑low prices remain sustainable at large scale?

What is already clear is that token pricing—once considered a relatively stable part of the AI stack—is becoming a competitive weapon. And if DeepSeek’s pricing holds, the cost of building large‑scale AI systems may fall far faster than many developers expected.

According to the official DeepSeek API documentation, once the promotion ends the model’s API pricing will be permanently adjusted to one‑quarter of its original level, effectively making the discounted pricing the new standard rate.

Before the discount, the listed base price was roughly $1.74 per million input tokens and $3.48 per million output tokens, meaning the current pricing represents a lasting step‑change rather than a short‑term marketing promotion.

How it compares with GPT‑5.5 and Claude Opus 4.7

DeepSeek’s pricing stands out most clearly when compared with other frontier AI models.

OpenAI GPT‑5.5

• Input: $5.00 per million tokens (short context)
• Output: $30.00 per million tokens

Anthropic Claude Opus 4.7

• Input: $5.00 per million tokens
• Output: $25.00 per million tokens

DeepSeek V4‑Pro

• Input: $0.435 per million tokens
• Output: $0.87 per million tokens

That means:

• Input tokens are roughly 91% cheaper than GPT‑5.5 or Claude Opus 4.7.
• Output tokens are roughly 96–97% cheaper than those flagship models.

For developers running high‑volume workloads—such as coding assistants, agents, document analysis pipelines, or customer‑support systems—token pricing often dominates total operating costs. A difference of this magnitude can change which architectures are economically viable.

Large‑context capabilities

Beyond pricing, V4‑Pro also competes technically with frontier models.

Key specifications include:

• 1‑million‑token context window
• Maximum output of up to 384,000 tokens
• Support for structured JSON output and tool/function calling

The model is built using a Mixture‑of‑Experts (MoE) architecture with roughly 1.6 trillion total parameters and about 49 billion active parameters per inference step, enabling high capacity without proportionally higher compute costs.

Large context windows are especially useful for:

• long‑document analysis
• codebase‑scale reasoning
• agent workflows with persistent context
• retrieval‑augmented generation with large knowledge bases

Combined with extremely low token pricing, these capabilities enable use cases that would be prohibitively expensive with higher‑priced models.

Why this matters for developers

For many teams, inference cost—not model capability—is the limiting factor when deploying AI at scale. Ultra‑low token prices allow developers to:

• run larger prompts and longer conversations
• generate longer outputs without cost spikes
• process huge document sets or codebases
• build multi‑agent pipelines with many model calls

In practical terms, the difference between $0.87 and $25 per million output tokens can turn previously experimental workflows into economically viable products.

The broader AI pricing war

DeepSeek’s strategy reflects a larger shift underway in the AI industry.

Historically, frontier models were priced in the range of several dollars to tens of dollars per million tokens, as seen with GPT‑5.5 and Claude Opus 4.7.

DeepSeek instead pursues a volume‑driven pricing model, offering dramatically cheaper inference while maintaining competitive capability. Analysts note that its models have repeatedly undercut incumbent providers by large margins, helping push overall AI costs downward across the market.

This approach mirrors dynamics seen in cloud computing and GPUs: once a vendor proves high‑performance systems can run much cheaper, the entire market is forced to respond.

What to watch next

DeepSeek’s pricing move raises several strategic questions for the AI ecosystem:

• Will competing labs lower API prices to remain competitive?
• Will cheaper inference accelerate the shift toward agent‑based applications?
• Can ultra‑low prices remain sustainable at large scale?

What is already clear is that token pricing—once considered a relatively stable part of the AI stack—is becoming a competitive weapon. And if DeepSeek’s pricing holds, the cost of building large‑scale AI systems may fall far faster than many developers expected.