DeepSeek Makes 75% V4 Pro Price Cut Permanent, Deepening AI Price War
Summary
- • DeepSeek permanently cut V4 Pro prices 75%, to $0.87 per million output tokens
- • New pricing undercuts GPT-5, Claude Opus 4.7, and Gemini across input and output costs
- • Price cut reflects architectural efficiency gains, not a promotional discount
- • V4 Pro is open source with 1M context window; enterprises can self-host to cut costs further
Details
DeepSeek V4 Pro pricing permanently cut 75%, from $3.48 to $0.87 per million output tokens
Input pricing also cut from $0.0145 to $0.003625 per million tokens (cache hit tier). The cut was originally promotional with a May 31 expiry; making it permanent signals a structural pricing shift, not a temporary competitive move.
DeepSeek V4 Pro now undercuts every major Western frontier model on both input and output token pricing
GPT-5 costs $2.50 input / $10 output per million tokens. Claude Opus 4.7 costs $5 input / $25 output. Even Google's budget-tier Gemini 3.5 Flash at $0.15 input / $0.60 output sits above DeepSeek's new ceiling. The gap is largest against frontier reasoning models used for demanding enterprise workloads.
Price cut attributed to architectural efficiency: one-quarter compute and one-tenth memory footprint vs. predecessor at long context
According to Greyhound Research analyst Sanchit Vir Gogia, the cut is an efficiency gain being passed through rather than a discount — V4 Pro was engineered specifically to reduce long-context inference costs, distinguishing it from margin-sacrificing promotional pricing.
V4 Pro supports a 1 million token context window and is fully open source
The 1M context window positions V4 Pro for document analysis, legal review, and codebase comprehension — workloads where token costs compound at scale. Being open source allows enterprises to run local deployments, potentially eliminating external API costs entirely.
Salesforce projects $300M in Anthropic token spending in 2026 — equivalent workloads on DeepSeek would cost a fraction of that
The Salesforce figure illustrates the scale of enterprise exposure to token pricing. If large customers begin routing lower-complexity tasks to DeepSeek, Anthropic's token volumes may hold while revenue per token declines, pressuring the economics behind its valuation trajectory.
Anthropic annualized revenue grew from $9B to $30B between late 2025 and early April 2026, driven by Claude Code
That growth trajectory is now under pricing pressure from DeepSeek. If enterprises tier their workloads — using DeepSeek for routine tasks and Claude only for high-stakes reasoning — Anthropic faces revenue-per-token compression even if total usage remains stable.
Anthropic accused DeepSeek of distillation attacks — allegedly training on Claude outputs to improve its own models
DeepSeek has not publicly responded to the accusation in detail. If substantiated, distillation attacks would raise significant IP and licensing questions. The allegation complicates enterprise procurement decisions involving both companies.
Analyst: DeepSeek V4 Pro has closed the performance gap on math and reasoning but lags on ecosystem and hyperscaler integrations
Neil Shah of Counterpoint Research characterizes V4 Pro as a formidable alternative on raw capabilities, while flagging primary limitations are ecosystem-level — global support, IP provenance, and deep hyperscaler integrations — not intelligence-level.
DeepSeek explicitly prioritizing market share over per-unit revenue with this permanent cut
The company framed V4 as ushering in an 'era of cost-effective 1M context length,' positioning it as the default for high-token-volume applications. The strategy mirrors classic platform land-and-expand plays: capture volume at low margin, then build switching costs through ecosystem and tooling.
OpenAI pivoting toward consumer platform revenue as API token revenue alone may not sustain its $852B valuation
DeepSeek's permanent price cut accelerates an industry-wide commoditization trend already compressing API margins throughout 2026. Google has repeatedly cut Gemini prices to compete with open-weight models. The structural shift suggests frontier AI API pricing may continue declining regardless of DeepSeek's individual moves.
Financials = pricing and revenue data, Market Impact = competitive effects, Tech Info = architecture and capabilities, Product Launch = new model features, Industry Update = company performance metrics, Legal = IP and compliance issues, Insight = analyst or expert interpretation, Strategy = business positioning
What This Means
DeepSeek's decision to lock in its 75% price cut permanently marks a structural inflection point in enterprise AI economics: what was a promotional pressure tactic is now a standing market reality that every major AI vendor must respond to or justify ignoring. For enterprise buyers, the immediate question is whether to tier workloads — routing high-volume, lower-stakes tasks to DeepSeek while preserving premium Western models for sensitive or compliance-critical use cases — and that tiering decision alone could materially reshape token revenue for Anthropic, OpenAI, and Google. The deeper implication is that frontier AI API pricing is on a one-way trajectory toward commoditization, and the competitive moats that will matter long-term are not price or context length but ecosystem depth, compliance posture, and integration into enterprise toolchains.
Sentiment
Broadly excited about commoditization and tiered workflows, with skepticism on sustainability
“Chinese LLMs just took the price war to an extreme. DeepSeek slashed its flagship API by 75%, bringing it close to “electricity cost.” ... We may be heading toward clearer stratification: Western models continue competing at the “ceiling,” while Chinese models dominate the “ground floor” and “middle layer.””
“Chinese AI labs are turning price into a weapon. DeepSeek just made its 75% price cut on V4-Pro permanent. The AI race is not only about the smartest model anymore. It is about who can make intelligence cheap enough to use everywhere. Good enough + cheap may beat best + expensive in many real workflows.”
“DeepSeek making the 75% V4-Pro API discount permanent is a loud move. If the model is good enough for tool calling and reasoning, cheaper tokens immediately change what builders are willing to automate.”
“Called it a discount. Turns out it was the price. #DeepSeek just made the 75% V4-Pro cut permanent. The token economics I wrote about 5 days ago aren't a promo window anymore – they're the floor.”
“DeepSeek making the 75% V4-Pro price cut permanent isn't a stunt. Frontier inference is converging on commodity pricing faster than the US labs planned for. The premium-token playbook is breaking. Only labs with real product gravity survive the squeeze.”
Split
~60/40 optimistic on developer/enterprise benefits vs. concerns over Chinese model sustainability and long-term Western moats; main split is tiered hybrid usage vs. full commoditization fears.