DeepSeek Hikes V4 API Pricing to Manage High Demand
DeepSeek is raising API prices for its V4 models by up to 1,100 percent to manage capacity constraints, forcing developers to strategically schedule workloads to avoid peak rates.
Chinese AI provider DeepSeek is ending its flat-rate pricing by introducing peak and off-peak rates for its V4 model family, effective August 16. The change comes as surging demand strains compute capacity. Under the new system, V4-Flash, currently in beta, will see input rates for cache misses rise from a flat $0.14 per million tokens to $0.22 off-peak and $0.44 at peak. Flash output rates will jump from $0.28 per million tokens to $0.66 off-peak and $1.32 at peak. These represent increases of 57% to 214% for inputs, and 136% to 371% for outputs.
The newly generally available V4-Pro model is also seeing price hikes. Its input rates for cache misses will rise from $0.435 per million tokens to $0.66 off-peak and $1.32 at peak, an increase of 51% to 203%. Pro output rates will climb from $0.87 to $1.98 off-peak and $3.96 at peak, a 127% to 355% increase. Furthermore, inputs with cache hits will experience the most dramatic hikes, surging between 52% and 1,100%. This directly targets DeepSeek's generous 98% cache-hit discount, which far exceeds the 90% industry average.
These changes alter the competitive landscape against OpenAI. At peak, V4-Flash loses its cost advantage over OpenAI 5.6 Luna, which recently saw an 80% off-peak price drop. Off-peak, V4-Flash is marginally more expensive on inputs and 45% cheaper on outputs than Luna. At peak, V4-Pro runs close to five times Luna's price for coding-agent workloads. However, V4-Pro retains its price advantage over OpenAI's mid-tier reasoning model, Terra, and the GPT-5.6 Sol model. DeepSeek's 98% cache-hit discount previously kept its cost per task 60% below Luna, but Flash's edge over Luna will now shrink from sevenfold to threefold off-peak, and 1.4 times at peak.
For developers, this shift turns timing into a critical economic variable. Because off-peak rates apply for 17 out of every 24 hours, Western buyers can mitigate costs by scheduling non-urgent workloads during these cheaper windows. Both V4-Pro and V4-Flash feature new flexible reasoning capabilities and chain-of-thought thinking modes. To manage these rising costs, practitioners must adopt sophisticated model routing and workload scheduling, treating model intelligence as a flexible commodity rather than a fixed dependency.
This is our own summary of reporting by Computerworld AI
