From Temporary Promotion to Structural DeepSeek API Pricing Shift
DeepSeek has converted its 75 percent discount on the DeepSeek-V4-Pro model into permanent API pricing, turning what looked like a short-lived promotion into a lasting structural shift in AI model cost reduction. The company’s pricing page now lists V4-Pro at one quarter of its original launch rate: USD 0.435 (approx. RM2.0) per million uncached input tokens and USD 0.87 (approx. RM4.0) per million output tokens, down from crossed-out reference prices of USD 1.74 and USD 3.48. Cached input now sits at USD 0.003625 (approx. RM0.02) per million tokens after the same 75 percent reduction. V4-Flash comes in even lower at USD 0.14 (approx. RM0.65) per million input tokens and USD 0.28 (approx. RM1.3) per million output tokens. By locking in the V4-Pro discount permanent instead of letting it expire, DeepSeek is signaling that aggressive, affordable AI development pricing is now central to its strategy, not a marketing experiment.
Why Long-Context and Cached Workloads Gain the Most
The new DeepSeek API pricing especially targets long-context AI workloads and heavy reuse of prompts, instructions, and reference material. V4-Pro and V4-Flash both support a one-million-token context window, allowing single requests to span long documents, complex multi-step instructions, or large codebases. DeepSeek has also cut input cache-hit prices across its model lineup to one tenth of launch pricing, and V4-Pro’s cached input is now USD 0.003625 (approx. RM0.02) per million tokens, while V4-Flash cache hits cost USD 0.0028 (approx. RM0.01). For support bots, coding assistants, RAG-style document search, and agentic systems that repeatedly reuse the same context, these lower cache rates dramatically reduce the marginal cost of every extra conversation or task. Instead of aggressively trimming prompts or summarizing away detail to save tokens, teams can keep more context inside the model while maintaining predictable budgets for high-volume, long-context AI operations.
Budget Predictability and New Room for Startup Margins
Making the V4-Pro discount permanent removes a looming cost reset and gives developers a stable baseline for financial planning. V4-Pro now sits at a quarter of its prior list price, at roughly USD 0.435 (approx. RM2.0) per million uncached input tokens and USD 0.87 (approx. RM4.0) per million output tokens, instead of reverting after May 31. For AI-native startups, this reliability matters: their products often look like software but behave like usage-based infrastructure, where every answer, generated report, or coding task carries a real token bill. Unstable pricing can wreck gross margins or force conservative rollout plans. With a standing lower rate, founders can design features, pricing tiers, and go-to-market strategies without modeling a sharp near-term jump in serving costs. That stability makes low-ticket, high-usage AI features more viable and allows finance teams to treat V4-Pro as a durable cost base instead of a temporary bargain.
Competitive Pressure on Premium AI Providers
DeepSeek’s decision clearly aims to undercut major AI API competitors and capture market share in budget-sensitive segments. V4-Pro, at USD 0.435 (approx. RM2.0) per million uncached input tokens and USD 0.87 (approx. RM4.0) per million output tokens, now looks unusually cheap in a market that has already been trending downward on pricing. For some workloads, the model may be 20 to 35 times cheaper than premium offerings from providers such as OpenAI, Anthropic, and Google, depending on prompt structure and output volumes. This changes comparison tables for buyers evaluating long-context coding tools, document-heavy applications, or agentic systems. Rivals must now justify substantially higher invoices or respond with their own reductions. DeepSeek is consciously trading margin for reach, leveraging a product ladder in which V4-Flash handles cheaper everyday workloads while V4-Pro focuses on complex, higher-value automation — potentially lowering the blended cost of an entire application portfolio.
How Affordable AI Development Changes Product Design Choices
Cheaper, predictable access to V4-Pro reshapes fundamental build-versus-buy decisions for developers and startups. Previously, teams weighing external APIs against fine-tuning open models or building narrow in-house systems had to account for high, uncertain token spend that could erode margins as usage grew. With DeepSeek’s permanent 75 percent cut, external APIs become more compelling for many use cases, especially when long context and rich interaction are critical. Startups can experiment with more generous context windows, richer system instructions, and multi-step agent workflows without immediately pricing themselves out of their own markets. Low-ticket products aimed at students, solo professionals, or very small businesses — segments traditionally squeezed by premium AI pricing — become more realistic. While reliability, latency, data policies, and trust still matter, the new economics encourage wider prototyping and faster iteration, accelerating affordable AI development and expanding who can sustainably ship AI-powered features at scale.