DeepSeek has launched its latest model, V4-Flash, once again pushing the cost competition in AI models to the forefront.
According to the latest calculations from research firm Artificial Analysis, the average cost per task completion for V4-Flash is approximately 3 US cents, making it the lowest among major global models. This is roughly equivalent to 1/105th of the cost of Anthropic's Claude Fable 5. This suggests that, for many general tasks, the marginal cost of deploying AI applications for enterprises could see further reductions, once again sparking discussion about the return on massive AI capital expenditures.
It is noteworthy that the release of V4-Flash comes at a time when DeepSeek is accelerating its commercialization process. Reports citing informed sources indicate that DeepSeek is preparing for a potential initial public offering (IPO). Its annualized recurring revenue (ARR) from its AI model API cloud services has reached between $400 million and $500 million, with a gross profit margin exceeding 50%.
The newly launched V4-Flash continues the company's consistent strategy of low cost and high efficiency, further strengthening its market competitiveness driven by price and performance.
V4-Flash "Cost Per Task" at Just 3 Cents, Leading Global Models in Cost-Effectiveness
According to data from Artificial Analysis, V4-Flash is priced at $0.14 per million input tokens and $0.28 per million output tokens.
However, the firm points out that, compared to simple list prices, the "Cost Per Task" provides a more accurate reflection of a model's actual usage cost. This metric takes into account the total number of input and output tokens a model consumes to complete the same task, making it a more meaningful reference.
Calculated using this metric, V4-Flash's average cost per task is about 3 cents, the lowest among major global models.
For comparison, Moonshot AI's Kimi K3 is approximately 86 cents, OpenAI's GPT-5.6 Sol is about $1.86, and Anthropic's Claude Fable 5 is roughly $3.15. V4-Flash's cost is equivalent to only about 1/105th of the latter.
Artificial Analysis further emphasizes that a model with a low per-token price could still have a higher final usage cost if it needs to generate more content or undergo more interaction steps to complete a task, compared to a more efficient model with a higher price. Therefore, the "Cost Per Task" is a better indicator of a model's true value than a single price metric.
Breaking Out with Low Prices, AI Competition Shifts from "Performance Battle" to "Value Battle"
Compared to performance rankings, V4-Flash has garnered more market attention for its cost advantage.
Over the past year, the focus of competition in the AI industry has gradually shifted from model capabilities to reasoning costs and commercialization efficiency. As enterprise AI applications scale up, model operating costs are becoming a key factor in procurement decisions.
When a model can complete a task for about 3 cents, while some high-end models cost several dollars, the overall usage cost for enterprises deploying high-frequency scenarios like customer service, office automation, and code assistance could see a dramatic reduction.
This once again touches on a persistent concern on Wall Street: whether massive AI capital investments will ultimately translate into sufficient commercial returns.
In early 2025, after DeepSeek released the R1 model, the market briefly reassessed the necessity of global AI infrastructure investment, causing significant volatility in tech stocks. Now, with V4-Flash once again using extremely low operating costs as its core selling point, it signals that competition in the AI industry is further shifting from "whose model is stronger" to "who can provide sufficiently good capabilities at a lower cost," and the discussion around AI investment returns is likely to heat up again.