DeepSeek raised prices for its flagship V4 AI models starting August 16, 2026, increasing the peak-hour rate for the V4-Flash model from $0.28 to $1.32 per million output tokens and the V4-Pro from $0.87 to $3.96 per million output tokens [1, 2, 3, 4, 5, 6]. The company also introduced differentiated peak and off-peak pricing, with off-peak rates set at 50% of peak prices to encourage users to schedule workloads outside congested times. "With the V4 lineup release, we’re updating our API pricing and introducing peak and off-peak rates. Off-peak rates are 50% lower than peak, enabling more flexible workload scheduling," DeepSeek stated [7].

These adjustments apply not only to output tokens but also to input tokens, with some prices increasing by up to 1100%. For example, peak-hour input tokens cached for V4-Pro saw increases from 0.025 RMB to 0.15 RMB, while output tokens went from 6 RMB to 13.5 RMB, reflecting the scale of these hikes [8, 9]. DeepSeek explained it aims to allocate resources more efficiently and encourage developers and enterprises to use AI models during less busy periods [1].

On August 13, DeepSeek launched the V4-Pro-0813 model, featuring improved AI agent capabilities and flexible "thought effort" settings (low, high, max) that adjust reasoning intensity based on task complexity [2, 7, 8, 5]. This model rollout coincided with the announcement of the new pricing structure.

Despite the sharp rise, DeepSeek’s pricing remains well below competitors. Anthropic’s Fable 5 charges $50 per million tokens, and Moonshot AI's Kimi K3 prices at $15, making DeepSeek’s new peak rates substantially more affordable [1, 3, 4, 6].

Financially, DeepSeek raised about $7.4 billion in its first external financing round in June 2026, amid talks of a potential $74 billion valuation. The company is preparing for an initial public offering as soon as this year [1, 2, 3, 5, 6].

The new pricing including peak and off-peak rates took effect Thursday at 1600 UTC and will shape how customers plan AI workloads going forward [1, 7, 3, 4, 9, 6].