DeepSeek V4-Flash Runs for $0.03 per Test — 100x Cheaper Than Claude Fable 5

For AI engineers and enterprise architects, inference cost is the silent killer of scalability. DeepSeek's V4-Flash delivers comparable intelligence to Google's Gemini at a per-task price that makes even basic on-device processing look expensive. This isn't just a new model release—it’s a signal that the unit economics of AI deployment are about to be rewritten.

Chinese AI startup DeepSeek has once again jolted the artificial intelligence landscape, this time with the release of its V4-Flash model, which benchmarks show is by far the cheapest among well-known large language models to run. Research firm Artificial Analysis published cost-per-test figures that place V4-Flash at just $0.03 on average, a staggering 105 times cheaper than Anthropic’s Claude Fable 5 ($3.15) and 62 times cheaper than OpenAI’s GPT-5.6 Sol ($1.86). The comparison even undercuts domestic rivals like Moonshot AI’s Kimi K3, which came in at $0.86 per test. The numbers reinforce DeepSeek’s reputation as the most aggressive price disruptor in generative AI, a strategy it famously deployed with its R1 reasoning model in early 2025, triggering a global tech stock selloff and forcing a re-evaluation of AI infrastructure spending.

“

Research firm Artificial Analysis published cost-per-test figures that place V4-Flash at just $0.03 on average, a staggering 105 times cheaper than Anthropic’s Claude Fable 5 ($3.15) and 62 times cheaper than OpenAI’s GPT-5.6 Sol ($1.86).

V4-Flash’s headline pricing is $0.14 per million input tokens and $0.28 per million output tokens. But the more enlightening metric is the all-in cost per completed task. Artificial Analysis measures this by accounting for the actual amounts of data a model must process and generate to finish a standardized evaluation, revealing that low headline rates can be deceptive if a model requires many more steps. On that real-world basis, V4-Flash’s efficiency is undeniable. It achieved a score of 50 out of 100 on the firm’s Intelligence Index, which aggregates nine benchmarks spanning coding, reasoning, and workplace-style assignments. That score ties it with Google’s Gemini model—a sign that extreme low cost need not come with catastrophic quality degradation, though it trails top-tier reasoning models in absolute performance.

The market context is one of intensifying price commoditization. U.S. hyperscalers and model developers have invested hundreds of billions of dollars into training and inference infrastructure, relying on premium API pricing to recoup costs. DeepSeek’s ultra-cheap offering challenges that model directly, echoing the post-2023 realization that inference efficiency can be dramatically improved through architectural innovations rather than brute-force compute. The V4-Flash model arrives at a time when the startup is reportedly preparing for a potential initial public offering, and its ability to demonstrate a sustainable low-cost advantage could be central to its valuation story. Yet DeepSeek is no longer the only Chinese player in town; it faces fierce competition from Moonshot, MiniMax, Z.AI, ByteDance, and Alibaba, all vying with U.S. titans for global enterprise adoption.