DeepSeek’s latest model has reset the economics of running frontier-grade AI. According to research firm Artificial Analysis, reported by Quartz on 3 August 2026, DeepSeek V4-Flash costs roughly 3 cents per test run — versus $3.15 for Anthropic’s Claude Fable 5 — making it by far the cheapest of the major models to operate.
The V4 family, tracked under releases such as DeepSeek-V4-Flash-0731, continues the Chinese lab’s strategy of pairing open weights with aggressive pricing. Open-weight availability means any developer can download, self-host, and fine-tune the model rather than paying per API call, and it narrows the functional gap with Western frontier systems at a fraction of the cost.
The pricing pressure is reshaping buyer decisions. For founders, freelancers, and enterprises running high-volume agentic workloads, the calculus is shifting from “best model wins” to “best price-to-task wins.” A model that is 100x cheaper to run changes which features are economically viable to ship.
The broader story is the open-weight split in the industry. More than 230 companies — including Nvidia, Microsoft, Meta, and Google — have publicly argued that open-weight AI should stay widely available, while Anthropic has taken the opposite stance. DeepSeek’s pricing is the commercial edge of that debate, and it is landing at a moment when the model race has become a speed race, a pricing war, and a distribution war at once.