DeepSeek V4 Is Now Up to 34x Cheaper Than GPT-5.5 and Claude, Here’s What It Means for Big Tech

DeepSeek V4 Undercuts GPT-5.5 & Claude by Up to 34x

What happens to a trillion-dollar industry when its core product suddenly costs a fraction of what it did? That is the question facing big tech after DeepSeek’s latest move. The Chinese lab’s new V4 model matches much of what GPT-5.5 and Claude Opus 4.7 can do, yet a permanent price cut in late May now undercuts them by up to 34 times on some workloads. For investors in the world’s biggest technology names, cheap AI has stopped being a theory and become a pricing reality they can no longer ignore.

What are the Best ASX Stocks to invest in right now?

Why DeepSeek V4 Is So Cheap, and Why That Matters

DeepSeek V4 is “open-weight”, which means anyone can download it and run it on their own servers rather than paying to use it through DeepSeek’s systems. It comes in two versions, a larger Pro model and a smaller, faster Flash model, both able to handle up to one million tokens of context, roughly a small book’s worth of text in a single prompt.

The pricing is where it gets interesting. DeepSeek launched V4 as a preview on 24 April 2026, and the rates tell the story. At standard baseline rates, V4 Pro costs around US$1.74 per million input tokens and US$3.48 per million output tokens. Compare that to GPT-5.5 at US$5 and US$30, Claude Opus 4.7 at US$5 and US$25, and Google’s Gemini 3.1 Pro at US$2 and US$12. That already puts DeepSeek’s baseline at roughly one-sixth the cost of the Western frontier.

Then DeepSeek went further. In late May, it made its 75% launch discount permanent, locking V4 Pro at US$0.435 per million input tokens and US$0.87 per million output tokens. That blows the “one-sixth” comparison out of the water: at the new rate, DeepSeek’s output costs roughly one-thirty-fourth of GPT-5.5 and around one-twenty-eighth of Claude Opus 4.7. The gap is no longer a discount; it is a different order of magnitude.

That may sound technical, but the point is simple. The basic building blocks of AI are becoming a commodity. When near-frontier intelligence is this cheap, it gets much harder for companies to charge premium prices just for access to a chatbot or text generator.

What This Means for the Tech Giants

Here’s the catch for investors. A large part of the lofty valuations behind names like Alphabet (NASDAQ: GOOGL) and Microsoft (NASDAQ: MSFT) rests on the assumption that AI software carries fat, durable profit margins. Cheap, capable open-weight models challenge that directly. If enterprises start routing everyday tasks to DeepSeek and reserve premium models only for the hardest jobs, revenue per token could fall even as overall usage climbs.

It’s worth keeping perspective, though. DeepSeek V4 is not the most capable model available. GPT-5.5 and Claude Opus 4.7 still lead on the toughest reasoning and coding tasks, with V4 sitting closer to the previous generation’s frontier. Anthropic has also publicly accused DeepSeek of using “distillation” to train on rival models, a dispute that remains unresolved.

For Nvidia (NASDAQ: NVDA), the picture is mixed. More efficient models need less computing power per task, which could soften chip demand in the short term. But cheaper AI also drives wider adoption, and the rush to build private data infrastructure to run these models locally could support hardware demand in the long term.

The bottom line is that the era of high-margin AI tokens looks like it is ending faster than many expected. For investors, that makes pricing power, not just model quality, the metric to watch from here.