Chinese AI developer DeepSeek has introduced V4-Flash, a new large language model that, according to independent research by Artificial Analysis, has the lowest operational cost among internationally known models. In tests cited by the firm, V4-Flash is more than a hundred times cheaper to run than Anthropic's Claude Fable 5.
DeepSeek unveiled V4-Flash on Friday as part of its push to strengthen market position amid intensifying competition in China’s AI sector.
Cost figures and practical implications
DeepSeek's published pricing charges $0.14 per million input tokens and $0.28 per million output tokens for V4-Flash. Artificial Analysis estimates that a typical test run on V4-Flash costs roughly $0.03.
By comparison, the same analysis reports average test costs for several other models as:
- Moonshot AI Kimi K3: $0.86
- OpenAI GPT-5.6 Sol: $1.86
- Anthropic Claude Fable 5: $3.15
Artificial Analysis notes that its comparison goes beyond list prices and also accounts for how much data a model needs to process and generate to complete a task, which it says gives a more accurate picture of real operational expenses.
Performance: comparable to some peers, behind the leaders
On Artificial Analysis’s intelligence index, DeepSeek V4-Flash scored 50 points—equal to Google Gemini 3.6 Flash and one point ahead of Meta Muse Spark 1.1 and Z.AI GLM-5.2. However, it trails Moonshot AI's Kimi K3 (57 points) and a group of higher-scoring models including Anthropic Claude Opus 5, Claude Fable 5 and OpenAI GPT-5.6, each of which scored at least nine points higher.
The picture that emerges is that V4-Flash offers a strong cost advantage that could be attractive for budget-sensitive deployment, while higher-performing (and costlier) models remain preferable for tasks that demand top-tier capability.
Market context and next steps
DeepSeek first attracted wide attention early in 2025 with its R1 model. Since then, a number of Chinese competitors—among them Moonshot AI, MiniMax, Z.AI, ByteDance and Alibaba—have rolled out significant improvements and are increasingly contesting global enterprise AI market share with U.S. technology firms.
DeepSeek said it is also preparing a higher-performance V4-Pro model, but has not announced a release date. Separately, Alibaba on Monday unveiled its largest and most advanced model to date, Qwen3.8-Max, which the company says places it at the forefront of recent Chinese developments.
Given current metrics, V4-Flash’s combination of very low operating cost and middling performance may make it an appealing option in scenarios where cost-efficiency outweighs the need for peak accuracy or capability.
