DeepSeek is doing what made it famous: undercutting everyone on price. The Chinese startup officially released V4-Flash on Friday, its latest bid to regain momentum after being overtaken by domestic rivals — and according to benchmark firm Artificial Analysis, it is by far the cheapest of all well-known models to run.
At $0.14 per million input tokens and $0.28 per million output tokens, V4-Flash averages just 3 cents per test — compared with 86 cents for Moonshot's Kimi K3, $1.86 for OpenAI's GPT-5.6 Sol and $3.15 for Anthropic's Claude Fable 5. That makes it roughly 105 times cheaper to run than Fable 5 on comparable tasks.
The per-task comparison matters because it accounts for how much data a model must process to finish a job: a low headline price means little if a model needs many more steps to answer. On Artificial Analysis' Intelligence Index — which combines nine benchmarks spanning coding, reasoning and workplace tasks — V4-Flash scored 50 out of 100, level with Google's Gemini 3.6 Flash and one point behind Meta's Muse Spark 1.1 and Z.AI's GLM-5.2. Frontier offerings from OpenAI and Anthropic scored nine or more points higher.
DeepSeek's R1 model triggered a global tech selloff in early 2025 by showing near-frontier performance at a fraction of the cost. The company — which sources say is preparing for a potential IPO — is also readying a more powerful V4-Pro, with no release date yet. The release lands amid a fierce race among Chinese AI labs, including Moonshot, MiniMax, Z.AI, ByteDance and Alibaba, to win businesses seeking cheap AI at scale; on Monday, Alibaba unveiled its largest model to date, Qwen3.8-Max.
