DeepSeek Locks In 75% Price Cut on Flagship V4-Pro Model
China's DeepSeek will make its 75% discount on V4-Pro permanent from 1 June, slashing flagship API costs to a quarter of launch rates. Indian builders stand to gain.
The News
China's DeepSeek will lock in a 75% reduction on its flagship deepseek-v4-pro model from 1 June 2026, turning what began as a promotional discount into a permanent floor for the company's most capable system. The change kicks in at 15:59 UTC on 31 May, when the existing temporary offer expires and base prices are formally rewritten to one-quarter of launch rates.
After the cutover, deepseek-v4-pro will cost $0.435 per million input tokens on a cache miss, $0.003625 per million on a cache hit, and $0.87 per million output tokens, down from $1.74, $0.0145 and $3.48 respectively. The cheaper sibling, deepseek-v4-flash, holds steady at $0.14 input and $0.28 output per million tokens.
The Hangzhou-based lab framed the move as a routine pricing adjustment rather than a fresh promotion, which means downstream products budgeting on the old sticker can rebuild their cost models with confidence. Bloomberg first reported the change on 23 May.
Why It Matters
DeepSeek has been the cheapest credible frontier-class option since its V3 release in late 2024, and the V4-Pro repricing widens that gap rather than closing it. At $0.87 per million output tokens, the model now sits roughly an order of magnitude below the headline pricing of comparable tiers from OpenAI's GPT-5.5 and Anthropic's Claude Opus 4, both of which still anchor near $10 to $15 per million output tokens for their flagship configurations.
The last time a Chinese lab forced a step-change in API economics was January 2025, when the original R1 release wiped close to $600 billion off Nvidia's market capitalisation in a single session and pushed every Western lab into a cost-disclosure cycle. This move is quieter but structurally more significant: a permanent price, not a promotion, signals that DeepSeek believes its inference stack can sustain these margins without venture support.
For application developers, that flips 'use the cheap model for everything' from a tactical hack into a viable production posture, and router products will likely recalibrate default model selections within days.
Indian Angle
Indian AI startups have been among the heaviest DeepSeek consumers outside China, partly because rupee economics make every order-of-magnitude pricing shift more material. At ₹83 to the dollar, deepseek-v4-pro now costs about ₹36 per million input tokens and ₹72 per million output tokens, which sits comfortably inside the unit-economics window that vernacular chatbots, document-extraction tools and small-business CRMs need to stay gross-margin positive.
Domestic foundation-model builders such as Sarvam, Krutrim and CoRover face a sharper competitive backdrop. Sarvam's enterprise pitch leans on sovereign-data positioning, but Indian IT-services buyers routinely benchmark domestic models against DeepSeek's price-per-token as a hurdle rate. A permanent V4-Pro floor makes that hurdle materially harder to clear.
Regulators may take notice too. MeitY's evolving guidance on cross-border AI inference, plus the data-localisation thread running through Digital India Act consultations, has already nudged some BFSI and healthcare buyers towards Indian or US-hosted models. Cheaper Chinese inference will reignite that procurement debate inside compliance teams, particularly at banks where Reserve Bank of India circulars treat model-hosting jurisdiction as a vendor-risk question.
FAQ
When does the new pricing take effect?
The permanent rates kick in at 15:59 UTC on 31 May 2026, which is 21:29 IST. The existing 75% promotional discount remains in force until that moment, so there is no gap between the cheaper sticker and the new official floor.
Which DeepSeek model is affected?
Only deepseek-v4-pro, the lab's flagship reasoning-grade model, sees the formal rewrite. The lighter deepseek-v4-flash stays at current rates, and cache-hit input pricing across both tiers benefited from a separate one-tenth reduction announced on 26 April.
How does this compare to OpenAI and Anthropic?
DeepSeek-V4-Pro at $0.87 per million output tokens prices roughly ten to fifteen times cheaper than the headline output rates for OpenAI's GPT-5.5 or Anthropic's Claude Opus 4 flagship tiers, although benchmark parity varies by task and language.
What does it mean for Indian developers?
Cost-sensitive applications, particularly Indic-language consumer products and SME automation, get more pricing headroom. Compliance-led buyers in BFSI and healthcare will still need to weigh hosting-jurisdiction questions before swapping models in production.
This story was reported by Bloomberg. Read the full original coverage at Bloomberg.