DeepSeek Slashes AI Model Prices by 17x: How a Chinese Startup Is Reshaping the Market

Chinese company DeepSeek has announced permanent price reductions for its flagship AI models, V4 Pro and V4 Flash, posing a serious challenge to the capital-intensive business models of Western labs like OpenAI and Anthropic. The cost of tokens for the V4 Pro model is now 7x lower on input and 17x lower on output compared to equivalents from Anthropic and OpenAI. The lightweight V4 Flash version is 10–25x cheaper than competing solutions, such as Claude Haiku.

These price changes were made possible by innovations in model architecture and hardware optimization. DeepSeek implemented unique cache compression mechanisms, drastically reducing data processing costs. When deployed in Chinese data centers, DeepSeek’s cache read costs are 87x lower than those of Western cloud providers. This has attracted major players: Xiaomi, for example, has adopted DeepSeek’s pricing model for its MiMo architecture.

DeepSeek’s models deliver performance comparable to Western counterparts. V4 Pro achieves 80.6% on the SWE-bench Verified coding benchmark and 87.5% on the advanced MMLU-Pro test. Both models are available with open weights under the MIT license, offering enterprises flexibility in deployment. The dual-model strategy optimizes costs: complex tasks run on V4 Pro, while fast operations use V4 Flash.

This move by DeepSeek increases pressure on Western companies, which have invested billions in infrastructure for general AI models. Open models are beginning to dominate the mass agent solutions segment, leaving closed models only for mission-critical tasks. For OpenAI, whose business model relies on mass API flows, this poses significant risks. For instance, Uber exhausted its annual token budget for Claude Code and Cursor in just four months, while Airbnb prefers cheaper alternatives like Alibaba’s Qwen.

Geopolitical barriers remain an obstacle to the widespread adoption of Chinese models in the West, particularly in regulated industries. However, for agile startups and teams unburdened by bureaucracy, infrastructure cost savings become a competitive advantage. According to OpenRouter, a platform for comparing AI models, DeepSeek V4 Flash now leads in processed token volume, surpassing Western competitors.