TL;DR
DeepSeek permanently cut V4 Pro prices by 75%, to $0.87 per million output tokens. It undercuts GPT-5, Gemini, and Claude.
DeepSeek has made permanent the 75% price discount on its flagship V4 Pro model. The promotion was originally scheduled to expire on 31 May. The Chinese AI startup’s pricing now ranges from $0.003625 to $0.87 per million tokens, down from $0.0145 to $3.48.
The price points are striking in context. OpenAI’s GPT-5 charges $2.50 per million input tokens and $10 per million output tokens. Anthropic’s Claude Opus 4.7 is priced at $5 input and $25 output.
Google’s Gemini 3.5 Flash, its cost-optimised model, charges $0.15 input and $0.60 output per million tokens. DeepSeek V4 Pro’s new permanent pricing sits below all of them. The gap is widest against the frontier reasoning models that enterprise customers rely on for demanding workloads.
The 💜 of EU tech
The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!
The decision to lock in the discount one month after launching the V4 models suggests DeepSeek is prioritising market share over per-unit revenue. The company described V4 as welcoming the “era of cost-effective 1M context length.” It is positioning its models as the default for applications that process large documents, codebases, or conversational histories where token costs compound fast.
For enterprise accounts consuming millions of tokens daily, the savings are material. Salesforce projects $300 million in Anthropic token spending this year. At DeepSeek’s new pricing, an equivalent volume would cost a fraction of that figure.
The question for enterprise buyers is whether DeepSeek’s model quality, reliability, and compliance posture justify the switch. The price advantage may be offset by the geopolitical and technical risks of routing sensitive workloads through a Chinese AI provider. That calculus varies by industry and by the sensitivity of the data involved.
The competitive dynamics are complicated by Anthropic’s public accusation that DeepSeek has engaged in “distillation attacks.” The allegation is that DeepSeek improperly trained on Claude’s responses to improve its own models. DeepSeek has not publicly addressed the accusation in detail.
If substantiated, it would mean that some of DeepSeek’s capability advantage was built on Anthropic’s research investment. The price differential would then reflect intellectual property arbitrage rather than engineering efficiency. The accusation remains unresolved.
Anthropic’s annualised revenue surged from $9 billion to $30 billion between the end of 2025 and early April 2026. That growth was driven largely by enterprise adoption of Claude Code. DeepSeek’s pricing pressure threatens the revenue-per-token economics that support Anthropic’s valuation trajectory.
If enterprise customers begin routing lower-complexity tasks to DeepSeek while reserving Claude for high-stakes reasoning, Anthropic’s token volume could hold while revenue per token declines. The broader AI pricing landscape has been moving toward commoditisation throughout 2026. Google has repeatedly cut Gemini prices to compete with open-weight models.
OpenAI’s pivot toward consumer platform features, including personal finance tools and advertising, reflects a recognition that API token revenue alone may not sustain its $852 billion valuation. DeepSeek’s permanent price cut accelerates a trend that was already compressing margins across the industry. The era of high-margin AI tokens may be ending faster than anyone expected.
DeepSeek V4 Pro supports a one-million-token context window at the new pricing. That makes it competitive for document analysis, legal review, and codebase comprehension. These are the long-context applications where input cost is the binding constraint on adoption.
The combination of frontier-adjacent capability and radically lower pricing creates a genuine dilemma for CTOs. The cheapest option is also the one with the most geopolitical complexity. It has the least transparency about training data provenance and an unresolved IP accusation from one of its most capable competitors.
DeepSeek’s strategy appears to be that price will win. Enough volume will flow to the cheapest capable model regardless of origin. The geopolitical concerns that constrain adoption in government and regulated industries will not prevent adoption in the broader market.
Whether that bet is correct depends on whether Western AI companies can close the price gap before DeepSeek closes the capability gap. The alternative is that the market bifurcates into a Western tier and a Chinese tier with fundamentally different economics. DeepSeek just made sure the gap between them got wider.
