DeepSeek V4 to Launch Mid-July with Peak Pricing Model

DeepSeek, the Chinese AI lab behind the open-weight DeepSeek family of models, has confirmed that the official version of DeepSeek V4 drops in mid-July 2026. The notice, sent via email to API subscribers, also detailed a significant shift in the company's billing structure. DeepSeek is introducing peak-load pricing, a move designed to better manage computational resources and keep the service humming during high-traffic periods.

Two Variants: Pro and Flash

DeepSeek V4 will come in two flavors: deepseek-v4-pro, the flagship heavyweight, and deepseek-v4-flash, a lighter, cheaper alternative. Both will have three pricing tiers, based on token type and whether the cache was hit.

For deepseek-v4-pro, here are the base (off-peak) rates:

• Input tokens (cache hit): $0.003625 per million tokens
• Input tokens (cache miss): $0.435 per million tokens
• Output tokens: $0.87 per million tokens

When peak hours kick in, those rates double. Input with cache hit jumps to $0.00725 per million tokens; input with cache miss climbs to $0.87 per million tokens; and output costs $1.74 per million tokens.

For deepseek-v4-flash, off-peak pricing is set at:

• Input tokens (cache hit): $0.0028 per million tokens
• Input tokens (cache miss): $0.14 per million tokens
• Output tokens: $0.28 per million tokens

Peak hours for Flash also mean a double, with output reaching $0.56 per million tokens.

When the Meter Runs High

Peak hours are defined in Coordinated Universal Time (UTC) as 01:00–04:00 and 06:00–10:00. In China Standard Time (UTC+8), that translates to 09:00–12:00 and 14:00–18:00, roughly the region's business day.

DeepSeek says the pricing update is meant to “better allocate resources and improve service stability.” The company will give users a 24-hour heads-up via email before the new rates kick in. Anyone still using the service after that is considered to have accepted the terms; users who don’t can cancel their subscription and get a refund.

The Strategy Behind the Shift

The move echoes what major cloud and AI outfits like OpenAI and Anthropic have already done: using tiered or capacity-based pricing to smooth demand fluctuations. Incentivizing off-peak usage helps DeepSeek ease congestion during high-traffic windows, a perennial headache for API providers.

DeepSeek has carved out a loyal following in the open-weight LLM world since the launch of DeepSeek-V2 and DeepSeek-Coder. With V4, the lab is taking a more commercial turn. The company hasn’t shared specific benchmarks yet, but the email promises “optimizations and performance improvements” over the current generation.

Developers have until mid-July to figure out whether the new pricing works for them. For high-volume inference workloads, the off-peak cache-hit rates, as low as $0.0028 per million input tokens for Flash, could be a bargain compared to rival APIs. But heavy users who hit the peak windows may find their bills running a lot higher.