DeepSeek just made its 75% price cut permanent
What started as a May promo is now the standing price. The cheapest serious model on the market just got cheaper for good.
Here are the numbers that matter. DeepSeek's V4-Pro model now costs $0.44 per million input tokens and $0.87 per million output tokens. GPT-5.5 charges $2.50 input and $15 output. That is more than 5x cheaper on input and roughly 17x cheaper on output, and as of this week it is not a sale anymore. It is the price.
DeepSeek ran a 75% promotional discount through May. The expectation was that it would lapse and prices would drift back up. They didn't. The Chinese lab has settled the model at the discounted rate and made it permanent, which changes the calculation for anyone building on top of large language models at scale.
For a startup pushing millions of tokens a day through an agentic workflow, the output-token gap is the one that hurts. A pipeline that costs $1,500 a day on GPT-5.5 output could run closer to $90 on V4-Pro. That is not a rounding error. That is the difference between a feature that ships and one that gets cut for being too expensive to serve.
There are caveats worth being honest about. DeepSeek is a Chinese lab, and plenty of enterprises have data-residency and audit requirements that rule it out regardless of price. V4-Pro is strong but it is not topping every benchmark, so for your hardest reasoning tasks you may still want a frontier model. The smart move most teams are landing on is a split: cheap model for the bulk, premium model for the few calls that genuinely need it.
The bigger story is what this does to the market. US business-spending indexes show DeepSeek climbing fast as companies swap pricey American APIs for cheaper alternatives on workloads where good enough is good enough. A permanent price this far below the frontier sets a floor that everyone else now has to argue against.
Which lands the pressure squarely on OpenAI. GPT-5.6 is expected in July. If it doesn't ship a clear, demonstrable quality lead, the pricing gap becomes very hard to defend to a finance team. The race in AI used to be startups undercutting each other. This is a well-resourced lab taking direct aim at the most valuable API business in the industry.