DeepSeek puts downward pressure on OpenAI, Anthropic, and Google pricing

DeepSeek reduced pricing for its latest AI model, V4-Pro, cutting rates by 75% for both input and output tokens. The maximum output token price fell from up to $3.48 per million to a maximum of $0.87 per million, with lower tiers also dropping substantially. DeepSeek stated the adjusted rates will largely remain after the temporary discount period ends. The change follows the recent introduction of the V4 series. Analysts attributed the price drop to improved efficiency, requiring less computing power and memory for complex tasks with large context windows. V4-Pro is positioned as open source, enabling local deployment and customization. Performance gains have narrowed gaps with Western models in some reasoning and computational tasks, while support, integrations, and major cloud adoption remain weaker.

"DeepSeek has drastically lowered the prices of its latest AI model, V4-Pro. The Chinese AI developer is cutting rates by 75 percent. This comes just a few weeks after the new V4 series was introduced. This further increases the pressure on competitors such as OpenAI, Anthropic, and Google. The price reduction applies to both input and output tokens."

"While using V4-Pro previously cost up to $3.48 per million output tokens, that rate is dropping to a maximum of $0.87. The lowest price tiers are also dropping significantly. According to DeepSeek, the adjusted rates will largely remain in place even after the temporary discount period ends."

"Analysts say the price cut primarily demonstrates how rapidly the costs of AI inference are falling. Sanchit Vir Gogia of Greyhound Research states in InfoWorld that DeepSeek has succeeded in making the model more efficient, requiring less computing power and memory for complex AI tasks with large context windows. The lower prices would therefore stem from technological optimizations rather than a temporary marketing campaign."

"DeepSeek unveiled its V4 models last month as the successor to the earlier R1 reasoning model. Like previous versions, the new models are also available as open source. This allows developers to run them locally and customize them for their own applications. According to Neil Shah of Counterpoint Research, DeepSeek has largely closed the performance gap with Western AI models in certain areas with V4-Pro."

Read at Techzine Global