CL-SN55 941 —
GPT-61S 890 —
GPT-56T 861 —
GPT-6S 837 -1.2%
CL-OP55X 831 -1.4%
MUSE-SPK 804 —
QWEN-38X 795 —
GPT-6A 782 —
CL-OP5H 724 -1.6%
CL-OP5X 722 -1.6%
KIMI-K3X 718 -0.1%
CL-OP46H 707 -1.7%
CL-FAB5H 705 -7.4%
CL-OP55H 705 -1.7%
CL-OP47H 696 -1.7%
GEM-37FH 661 -9.7%
GEM-38FH 656 -4.8%
GPT-56S 580 -0.5%
CL-OP47 545 -0.5%
INKL 531 —
GEM-3P 494 -0.2%
CL-OP46 491 —
GEM-31P 486 -0.4%
CL-OP48 484 —
GPT-6L 436 —
CL-SN55 941 —
GPT-61S 890 —
GPT-56T 861 —
GPT-6S 837 -1.2%
CL-OP55X 831 -1.4%
MUSE-SPK 804 —
QWEN-38X 795 —
GPT-6A 782 —
CL-OP5H 724 -1.6%
CL-OP5X 722 -1.6%
KIMI-K3X 718 -0.1%
CL-OP46H 707 -1.7%
CL-FAB5H 705 -7.4%
CL-OP55H 705 -1.7%
CL-OP47H 696 -1.7%
GEM-37FH 661 -9.7%
GEM-38FH 656 -4.8%
GPT-56S 580 -0.5%
CL-OP47 545 -0.5%
INKL 531 —
GEM-3P 494 -0.2%
CL-OP46 491 —
GEM-31P 486 -0.4%
CL-OP48 484 —
GPT-6L 436 —
← Back to feed

DeepSeek Doubles V4 Output Prices During Peak Beijing Hours — 6 Weeks After Locking In a 75% Cut

DeepSeek has introduced surge pricing on its V4 API. During peak Beijing business hours — 9am to noon and 2pm to 6pm China Standard Time — V4 Pro output rates double from 6 yuan to 12 yuan per million tokens, approximately $1.74 at current exchange rates.

The stated rationale is steadier service and better distribution of scarce compute resources during the hours when Chinese enterprise demand is highest.

Two Moves in Six Weeks

The timing is the story. In early June, DeepSeek permanently locked in a 75% reduction on V4 Pro pricing, dropping output to $0.87 per million tokens. That cut was positioned as permanent — locking in the economics of the V4 family ahead of V4.1.

Surge pricing does not reverse that cut. Off-peak rates stay at the permanently reduced level. But peak-hour prices on V4 Pro output are now roughly double the headline figure developers used to price their applications.

Any system built assuming a consistent $0.87/M output cost now pays twice that during the five hours of peak demand that overlap with Chinese business hours. For US-based developers running asynchronous batch workloads, the impact may be minimal. For teams running real-time inference on DeepSeek V4 during those hours, it is a 2x unit economics change.

What the Structure Creates

Demand-based pricing is standard in cloud infrastructure and consumer platforms. It is new in frontier AI APIs, where labs have competed primarily on flat per-token rates.

The pricing structure creates a clear routing incentive: batch jobs and asynchronous processing can be scheduled around the peak window. Real-time applications — customer-facing agents, live code assistants, synchronous multi-step workflows — cannot be deferred. Those use cases now pay a 2x premium precisely during the hours when Chinese enterprise demand is highest.

The practical effect is a bifurcated user base. Price-sensitive developers doing bulk processing will shift to off-peak scheduling. Teams with latency requirements will pay the surge rate or route to alternative providers during the peak window.

Background

DeepSeek’s aggressive pricing strategy has driven significant developer adoption. V4 doubled its OpenRouter token share in 60 days to reach the platform’s top position by traffic. At that scale, compute capacity management becomes a real commercial and engineering problem — and time-of-day pricing is the standard tool infrastructure operators use to flatten demand curves.

Whether other frontier AI providers follow with similar peak-hour pricing depends on whether the move improves margins and service reliability without pushing developers toward alternatives. The experiment is live.