DeepSeek will completely preserve the 75% low cost on its flagship V4-Pro mannequin that had been set to run out Could 31, the corporate mentioned Saturday. V4-Pro API output tokens now price $0.87 per million, down from $3.48 at launch one month in the past.
DeepSeek introduced Saturday that the 75% low cost on its flagship V4-Pro mannequin is now permanent, dropping API prices to as low as $0.0035 per million tokens and intensifying a price war with OpenAI, Anthropic, and Google.
The Chinese language AI startup had initially set the promotion to run out on Could 31, in line with its pricing web page. As an alternative, the corporate mentioned in a press release that V4-Pro API pricing would keep at 1 / 4 of launch charges indefinitely, with prices starting from 0.025 to six yuan ($0.003625 to $0.87) per million tokens relying on utilization sort.
Claude expenses 29 instances extra for a similar output token
With the new charge construction, the V4-Pro output tokens price $0.87 per million tokens. Claude Opus 4.7 prices $25 per million tokens. The price of GPT-5.5 is $30.
In line with CloudZero, Google’s Gemini 2.5 Professional prices $12. Different lower-cost Western LLMs, such as Claude Sonnet ($15) and GPT-4.1 ($8), are 9 instances to 17 instances extra
An utility producing 100 million output tokens per thirty days would price roughly $87 on V4-Pro on the new charge, versus $2,500 on Opus 4.7 or $3,000 on GPT-5.5, per MindStudio.
As Cryptopolitan earlier reported when V4 launched final month, DeepSeek mentioned the Professional model would initially price as much as 12 instances greater than the lighter Flash model because of “constraints in high-end compute capability.”
The corporate added that pricing would fall as soon as Huawei Ascend 950 supernodes shipped in bigger portions within the second half of 2026. The permanent cut arrived months earlier than that concentrate on.
DeepSeek didn’t say whether or not Huawei chips made the cut attainable
DeepSeek didn’t disclose whether or not the permanent price cut was because of the elevated provide of Huawei’s Ascend 950 chips.
Huawei’s AI chip gross sales have benefited from US export controls that stop Nvidia from promoting its most superior semiconductors in China. Separate restrictions on chipmaking gear have restricted Huawei’s capacity to scale Ascend manufacturing.
Huawei is focusing on 750,000 AI chip shipments in 2026, per trade estimates, although its complete manufacturing quantities to roughly 3% to five% of Nvidia’s mixed computing energy.
4 Chinese language chip corporations (Huawei Ascend, Cambricon, Hygon Info, and Moore Threads) confirmed their {hardware} labored with V4-Pro from launch day.
The V4 launch already triggered a selloff in Chinese language AI shares
In early April when V4 was launched, the share costs of some Chinese language synthetic intelligence corporations fell on the Hong Kong inventory alternate. Zhipu AI declined by about 8%-9% whereas MiniMax noticed a fall of seven%-8%.
In line with CNBC, Manycore Tech fell by 9%. Shares of chipmakers rose opposite to the pattern.
DeepSeek’s pricing technique has additionally drawn friction with Western rivals. Anthropic has beforehand accused the corporate of “distillation assaults” that improperly study from Claude’s fashions, per Engadget.
The White Home accused China final month of stealing US AI labs’ mental property. The Chinese language Embassy in Washington rejected these claims.
DeepSeek is concurrently in talks to lift exterior funding for the primary time, with Tencent and Alibaba amongst potential traders at a valuation above $20 billion, per Cryptopolitan.
“Crypto bots and DeFi brokers lastly have pricing that works”
Brokers that execute multi-step duties autonomously can burn by way of thousands and thousands of tokens per session.
The each day price for working an agent such as Claude Opus at a charge of $25 per million output tokens is tons of of {dollars}. DeepSeek V4-Pro at $0.87 is below $40.
Crypto buying and selling bots, DeFi watch brokers, code reviewing processes, and market evaluation functions all rely on the inference prices remaining beneath a threshold the place the worth of the output is bigger than the compute price.
The announcement by DeepSeek about its permanent pricing allows a class of agent functions to be possible on its platform that aren’t possible on Western platforms at present charge.
OpenAI has been reducing its costs all through the final 12 months. Anthropic launched Claude Haiku 4.5 for $1/$5 per million tokens. Google has rolled out Gemini Flash for a $0.30 enter price. Nevertheless, none of them may compete with DeepSeek’s pricing mannequin.
The V4-Pro is just not a funds mannequin. It has 1.6 trillion parameters, a 1 million token context window, and benchmarks that sit near GPT-5.4 on math and reasoning duties, per MindStudio evaluation.
Don’t simply learn crypto information. Perceive it. Subscribe to our e-newsletter. It is free.
