admin The Hangzhou-based company charges $0.14 per million input tokens and $0.28 per million output tokens for requests that do not use previously cached data. Its cache-hit price is substantially lower at $0.0028 per million input tokens, making the model particularly attractive for applications that repeatedly process similar documents, instructions or conversation histories.
Independent testing placed the average cost of running V4-Flash through a standard intelligence benchmark at about three US cents per task. That compared with roughly $0.86 for Moonshot AI’s Kimi K3, $1.86 for OpenAI’s GPT-5.6 Sol and $3.15 for Anthropic’s Claude Fable 5.
The comparison makes V4-Flash more than 100 times cheaper than Anthropic’s model under the benchmark’s test conditions. The gap is significant because inference — the process through which a trained model answers questions or performs tasks — is becoming one of the largest operating expenses for companies deploying AI at scale.
V4-Flash scored 50 on the test provider’s Intelligence Index, putting it level with Google’s Gemini 3.6 Flash. It remained behind several higher-performing frontier systems, indicating that its chief advantage lies in price-to-performance rather than outright leadership across every measure of reasoning, coding and knowledge.
Benchmark costs should not be treated as a universal estimate for business use. Actual spending depends on prompt length, response length, reasoning settings, caching, workload complexity and the number of times a model calls external tools. A system that produces longer answers or uses extended reasoning can consume substantially more tokens despite having a low advertised rate.
DeepSeek introduced V4-Flash as the smaller and faster member of its V4 model family. The company has also developed V4-Pro, which is designed for more demanding coding, reasoning and agent-based work. V4-Flash uses a sparse mixture-of-experts architecture, activating only part of its overall network for each token rather than using every parameter for every request.
That design reduces the amount of computation required during inference. DeepSeek has also used compressed memory techniques and specialised attention mechanisms to lower the cost of processing long sequences of text. Such improvements are increasingly important as developers build assistants capable of reviewing large code repositories, legal files, financial records and extended conversations.
The model’s pricing strengthens China’s position in the lower-cost segment of the global AI market. Moonshot AI, Alibaba, ByteDance, MiniMax and Zhipu AI are also competing through inexpensive application programming interfaces, open-weight releases and models optimised for coding or autonomous software agents.
Alibaba has expanded its Qwen range with higher-capability models, while Moonshot has promoted Kimi K3 as a cost-efficient alternative for developers requiring large context windows and agentic features. The growing number of capable systems is weakening the ability of any single provider to maintain premium pricing for routine tasks.
US developers continue to command strong demand for complex enterprise workloads, particularly where reliability, security controls, multimodal capability and integration with business software carry greater weight than token prices. Anthropic has focused on coding and enterprise safety, while OpenAI and Google are investing heavily in agents, search, productivity tools and multimodal services.
The widening price divide is encouraging companies to use several models rather than relying on one provider. AI routing platforms can direct basic classification, summarisation and customer-support requests to cheaper systems, reserving costly frontier models for difficult reasoning, research and software-engineering tasks.
Lower prices could expand AI adoption among smaller businesses that have been unable to justify high inference bills. Call centres, online retailers, financial technology companies and software developers may be able to automate larger volumes of work without committing to expensive computing infrastructure.
DeepSeek’s advance also raises questions about the economics of the AI industry. Leading technology groups are spending tens of billions of dollars on data centres, advanced semiconductors and electricity while the market price of model-generated intelligence continues to fall.
