Detailed Analysis
DeepSeek released its V4-Pro model on April 24–25, 2026, positioning it as a frontier-class open-source system with 1.6 trillion total parameters, 49 billion active parameters, and a 1 million token context window. The model is priced at $1.74 per million input tokens and $3.48 per million output tokens via API, with a smaller companion model, DeepSeek V4-Flash, available at just $0.14 input and $0.28 output per million tokens. NDTV Profit's framing of a "75% price cut" does not appear corroborated by available launch documentation; rather, the pricing reflects DeepSeek's foundational efficiency-first strategy enabled in part by scaling on Huawei Ascend 950 hardware. Benchmark results indicate V4-Pro leads open-weight models in math, STEM, and coding tasks, while approaching the performance of leading closed-source systems — though it processes output at a relatively modest 38 tokens per second.
The competitive pricing picture is stark when placed alongside Anthropic and OpenAI's offerings. Anthropic's Claude Haiku 4.5, positioned as a mid-tier model, runs $1 per million input tokens and $5 per million output tokens — meaning DeepSeek V4-Pro undercuts Claude on output pricing despite claiming comparable reasoning capability. At the top-tier level, Anthropic and OpenAI's flagship models are reported to command approximately $25–$30 per million output tokens, making DeepSeek V4-Pro roughly 88–90% cheaper on that dimension alone. Google's Gemini 3 Flash, at $0.50 input and $3.00 output per million tokens, sits closer to DeepSeek's range but still above it. The pattern illustrates that DeepSeek is not merely competing on price at the budget end — it is undercutting premium-tier pricing with a model claiming premium-tier performance.
This development carries significant strategic weight for the broader AI industry. While OpenAI and Anthropic have historically trended toward pricing increases alongside capability upgrades — reflecting heavy infrastructure investment and growing enterprise demand — DeepSeek has pursued the inverse model, treating aggressive affordability as a primary competitive lever rather than an afterthought. The open-sourcing of V4-Pro further amplifies its market impact, as it enables community deployment and fine-tuning independent of DeepSeek's own API, a pathway unavailable with closed systems like Claude or GPT. Reports suggest further price reductions on V4-Pro are anticipated later in 2026, which would deepen competitive pressure on Western labs already navigating tension between capability investment costs and market pricing expectations.
The broader trend this episode reflects is the accelerating commoditization of frontier-level AI inference. DeepSeek's V3 generation earlier established that Chinese open-source labs could match closed-source performance benchmarks at a fraction of the cost; V4-Pro advances that thesis with a model that now credibly competes in reasoning and agent-task categories previously dominated by Anthropic's Claude 3 and OpenAI's GPT-5 family. For Anthropic specifically, the pressure is nuanced: its brand has been built substantially on safety research and enterprise trust, not price leadership. However, as raw performance parity becomes an increasingly plausible claim from open-source competitors, the pricing differential between Claude's API tiers and models like DeepSeek V4-Pro becomes a harder sell to cost-sensitive developers and startups. The comparison also underscores the growing relevance of hardware sovereignty — DeepSeek's ability to scale on domestically produced Huawei chips, despite U.S. export controls on advanced Nvidia hardware, signals that the infrastructure assumptions underlying Western AI pricing strategies may be less durable than previously assumed.
Read original article →