DeepSeek Locks In 75% Forever: V4-Pro Becomes the Cheapest Frontier AI API on the Market

News Summary

DeepSeek, the AI research lab behind one of the world's most cost-competitive large language models, announced on May 22, 2026 (UTC) that its 75% discount on the V4-Pro API is being made permanent — a decision that marks a new phase in the global AI pricing competition and dramatically lowers the barrier to entry for developers building on frontier-class models.

Background: From Promotion to Policy

When DeepSeek launched the V4-Pro model in late April 2026, it introduced an introductory 75% discount on API access as a limited-time promotion. At that time, the discount was scheduled to expire on May 31, 2026 at 15:59 UTC. Less than a month later, on May 22, 2026 (UTC), DeepSeek updated its official API documentation and issued a public statement confirming that the discounted price would become the new permanent standard pricing, effective after the promotional window closes.

The decision came roughly one month after the V4 generation's debut and surprised many observers who expected prices to revert to their original levels before any longer-term repricing was announced.

New Pricing Breakdown

The permanent pricing for DeepSeek-V4-Pro positions the model as one of the most affordable frontier AI APIs currently available:

• Cache-hit input tokens: $0.003625 per million tokens (down from $0.0145)
• Cache-miss input tokens: $0.435 per million tokens (down from $1.74)
• Output tokens: $0.87 per million tokens (down from $3.48)

In Chinese yuan terms, this translates to a range of 0.025 to 6 yuan per million tokens, compared to the former range of 0.1 to 24 yuan. The across-the-board reduction represents a precise 75% cut at every pricing tier. According to analysis published by The Decoder on May 23, 2026 (Eastern Time), DeepSeek's output token pricing now sits at least 34 times below that of OpenAI's GPT-5.5 at comparable capability levels.

Technology Behind the Pricing Shift

A key technical factor underlying the price reduction is the DeepSeek V4 series' optimization for Huawei's Ascend AI accelerator hardware. Unlike previous DeepSeek generations that ran primarily on Nvidia GPUs, the V4 family was engineered from the ground up to leverage the Ascend 950 and 950PR AI supernode architectures.

When V4 was initially released in April 2026, DeepSeek stated that API pricing was expected to decrease sharply once Huawei Ascend 950 supernodes became available in large quantities, which the company projected would occur in the second half of 2026. The decision to make the discount permanent ahead of that supply ramp suggests that DeepSeek has gained sufficient confidence in its infrastructure cost trajectory to commit to the lower price tier now.

DeepSeek has not publicly disclosed the exact relationship between Ascend hardware availability and the permanent pricing decision, but the timing aligns with broader reports of increased Huawei Ascend 950 deployment within the AI industry.

Competitive Landscape

The permanent price cut places DeepSeek-V4-Pro in a uniquely advantageous position relative to its peer models. At the new rates, V4-Pro's pricing undercuts leading alternatives across the board:

• OpenAI GPT-5: Significantly higher output token costs
• Anthropic Claude Opus 4.7: Higher per-token pricing at comparable reasoning tiers
• Google Gemini 3.5 Flash: DeepSeek-V4-Pro's output pricing remains lower even against Flash-tier models

South China Morning Post reported on May 24, 2026 (Eastern Time) that DeepSeek-V4-Pro topped global "bang-for-buck" rankings following the permanent price announcement, based on benchmark performance relative to per-token cost.

Developer and Industry Impact

For developers and enterprises, the permanent nature of the discount removes a major planning uncertainty. Teams that had been evaluating DeepSeek-V4-Pro as a cost-saving alternative but were hesitant due to the temporary nature of the promotional pricing can now build production systems with greater confidence in cost stability.

Bloomberg reported on May 23, 2026 (Eastern Time) that the move is widely being interpreted as an escalation in the ongoing AI price war. In recent months, several major AI providers have reduced API prices in response to competitive pressure. DeepSeek's move to permanently lock in a 75% reduction on a flagship model raises the bar for what developers expect from frontier AI pricing.

The announcement is also notable as a signal for the broader AI infrastructure economy. If high-performance inference at sub-dollar-per-million-token pricing becomes the market norm, it may accelerate adoption of AI in cost-sensitive sectors such as education technology, developer tools, and small-business applications.

What Comes Next

DeepSeek has indicated that the official pricing adjustment will take effect after the promotional period concludes on May 31, 2026 at 15:59 UTC. After that date, the permanently reduced prices will be the standard rates in the API documentation. Developers currently using the promotional pricing will experience no interruption or change in their billing rates.

The company has not announced further pricing changes or additional model releases in conjunction with this decision, though speculation in the developer community continues about when V4-Pro will see a successor and whether similar permanent discounting strategies will apply to future model generations.