DeepSeek Makes 75% Price Cut on V4-Pro API Permanent

By Hector Herrera | May 23, 2026 | Business

DeepSeek is permanently locking in a 75% discount on its flagship V4-Pro API, setting new base prices that are a fraction of what U.S. competitors charge at comparable capability levels. The move signals that Chinese AI firms aren't treating low pricing as a promotional tactic — they're treating it as a market structure.

The cut was first offered as a limited-time promotion set to expire May 31. Bloomberg reported Friday that DeepSeek has decided to make those rates permanent before the deadline arrives.

The Numbers

The new permanent pricing for DeepSeek V4-Pro:

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

That's roughly one-quarter of the model's original list price. For context, OpenAI's GPT-4o currently runs at $2.50 per million input tokens and $10.00 per million output tokens at standard rates. Anthropic's Claude Opus 4.6 sits at $15.00 per million input and $75.00 per million output at the high end.

V4-Pro is DeepSeek's most capable production model, positioned to compete directly with those flagship offerings on reasoning and coding tasks.

Why It Matters

This changes the floor for API pricing. When a competitive flagship model — not a stripped-down distillation — is available for under $1 per million output tokens, every pricing conversation in the industry shifts.

Developers building AI-powered products do constant cost modeling. A 75% permanent reduction in inference cost doesn't just affect the economics of existing apps; it makes a new class of applications viable that weren't before. High-volume use cases — customer service agents processing thousands of calls, document analysis pipelines, real-time translation — all become dramatically cheaper to run.

The pressure lands hardest on mid-tier U.S. providers who compete on price rather than capability. Hyperscalers like OpenAI and Anthropic can absorb some pressure because they have enterprise relationships, regulatory compliance positioning, and U.S.-hosted infrastructure that some buyers require. Smaller API resellers and inference providers have less cushion.

The Competitive Context

DeepSeek's pricing strategy fits a pattern. Chinese AI firms — including Alibaba's Qwen and Baidu's ERNIE — have been systematically undercutting U.S. providers on API access, often by factors of 5x to 10x. The theory: capture global developer mindshare now, build platform dependency, and monetize later.

The strategy is working with cost-sensitive developer segments. A growing number of startups and open-source projects have migrated inference workloads to DeepSeek's API for exactly this reason.

There's a counterargument from enterprise buyers: data residency and geopolitical risk matter. U.S. companies processing sensitive customer data, operating in regulated industries, or subject to government contracting requirements often can't or won't use Chinese-hosted AI infrastructure regardless of price. That segment of the market is structurally ring-fenced from DeepSeek's pricing pressure.

But that's not the whole market. The global developer population — especially outside the U.S. and Europe — doesn't carry those constraints in the same way.

What to Watch

The May 31 date was the original promotional deadline. By locking in prices before that date, DeepSeek is sending a signal to developers who were waiting to commit: the price is real, plan around it. Watch whether OpenAI, Google, or Anthropic respond with price adjustments in the next 30 to 60 days — and whether any U.S. firms start tiering their offerings more aggressively to compete at the low end without cutting flagship rates.

Sources: Bloomberg