DeepSeek's V4-Flash is 100x cheaper than Claude Fable 5 on AI benchmarks

DeepSeek, the Chinese startup that shook global tech markets in early 2025 with its R1 model, is back with V4-Flash — a version of its flagship AI that the research firm Artificial Analysis says is the cheapest to run among well-known models worldwide. The pricing is striking: $0.14 per million input tokens and $0.28 per million output tokens. But more importantly, when measured by real-world task cost, V4-Flash averages just 3 cents per test. For comparison, Moonshot AI's Kimi K3 costs 86 cents, OpenAI's GPT-5.6 Sol costs $1.86, and Anthropic's Claude Fable 5 costs $3.15 per test. That means V4-Flash is roughly 100x cheaper than Anthropic's premium model on a task-adjusted basis, which accounts for how much data each model must process to generate an answer.

V4-Flash's low cost doesn't come without tradeoffs. It scores 50 out of 100 on Artificial Analysis' Intelligence Index, which blends nine benchmarks covering coding, reasoning, and workplace-style tasks. That's the same score as Google's Gemini 3.6 Flash, one point behind Meta's Muse Spark 1.1 and Z.AI's GLM-5.2, and well below Moonshot's Kimi K3 at 57 or Anthropic's Claude Opus 5 and GPT-5.6, which score nine or more points higher. Still, for price-sensitive enterprises, V4-Flash offers a compelling value proposition. DeepSeek is also reportedly preparing a more powerful V4-Pro model, though no release date has been announced. This move intensifies competition in the low-cost AI tier, where Chinese rivals like Alibaba (with its new Qwen3.8-Max) are also pushing aggressive pricing to capture global adoption.