DeepSeek AI pricing OpenAI Anthropic token costs

DeepSeek's 75% Price Cut Is Now Permanent. That's a 34x Gap on Output Tokens Versus GPT-5.5.

25 May 2026

Deepseek has made its 75% price discount on its flagship model, Deepseek V4 Pro, permanent, with output tokens now priced at just $0.87 per million — roughly 34 times cheaper than GPT-5.5's $30 per million. While Deepseek V4 Pro lags behind top frontier models like GPT-5.5 and Opus 4.7 in raw performance, the dramatic price gap makes it particularly attractive for token-heavy applications like agentic AI systems. However, raw token pricing doesn't tell the whole story, as token consumption per task also significantly affects real-world costs.

DeepSeek has quietly won a pricing war that Western labs probably didn't want to fight. The Chinese AI company has made its temporary 75% discount on DeepSeek V4 Pro permanent, scrapping the original May 31 expiry date and cementing what is now a staggering cost advantage over OpenAI and Anthropic.

The numbers are blunt. V4 Pro now costs $0.435 per million input tokens and $0.87 per million output tokens. GPT-5.5 charges $5 and $30 respectively. That's roughly 11.5 times cheaper on input and 34.5 times cheaper on output. If you're hitting GPT-5.5's long context pricing above 272K tokens, DeepSeek looks even better, around 23x cheaper on input and over 50x on output. DeepSeek V4 Flash undercuts V4 Pro further still, at $0.14 input and $0.28 output.

Both DeepSeek models support a one million token context window with up to 384,000 output tokens, and the company has been sensible enough to support both OpenAI and Anthropic API formats. Switching costs for developers are low, which is rather the point.

Of course, raw token pricing is only part of the story. Token efficiency matters enormously in practice. A model that costs less per token but burns through three times as many to complete the same task isn't actually cheaper. Google's Gemini Flash 3.5 is a good case study here: cheaper headline pricing than its predecessor, but higher token consumption that often cancels out the apparent saving. Anthropic's Opus 4.7 looked affordable compared to GPT-5.5 on paper too, yet in practice both ended up costing users somewhere between 30 and 90 percent more than the models they replaced.

DeepSeek V4 Pro is not in the same performance bracket as GPT-5.5 or Opus 4.7, and anyone claiming otherwise probably has a vested interest. How large the gap is depends heavily on the task, and benchmark scores are notoriously unreliable guides to real-world usefulness. But the price differential is so extreme that performance doesn't have to be equal to make DeepSeek the rational choice for a large number of workloads, particularly agentic pipelines that can consume tokens at a ferocious rate.

This is where the strategic picture gets interesting. AI spending is growing, ROI remains stubbornly difficult to quantify, and companies are understandably getting twitchy about the bills. The question many are now asking is not 'what's the best model?' but 'what's the cheapest model that won't embarrass us?' DeepSeek is positioning itself as the obvious answer.

DeepSeek is reportedly entering its first funding round, but it operates under nowhere near the financial pressure that OpenAI and Anthropic face, both of which are eyeing IPOs and carrying significant cost structures. Sustained low pricing is a much more credible long-term strategy for DeepSeek than it would be for its Western rivals. Whether that changes how seriously those rivals take the threat is another matter entirely.