DeepSeek's V4-Pro and V4-Flash launch with peak-hour price hikes and 1M context
Twice the API cost during peak hours for DeepSeek's new V4 models coming mid-July
DeepSeek has confirmed the official launch of its V4-Pro and V4-Flash models for mid-July, delivering a significant upgrade over the current generation. Both models feature a 1-million-token context window — enabling massive document processing — along with improved agentic capabilities (AI that can autonomously execute tasks), stronger mathematical reasoning, and code generation tailored for complex software engineering workflows. These enhancements position DeepSeek to compete directly with other frontier models like GPT-4o and Claude 3.5 in enterprise and developer environments.
Alongside the launch, DeepSeek introduced a new API pricing model that doubles the cost of tokens during peak hours (9am-12pm and 2pm-6pm daily). This dynamic pricing approach is designed to spread demand and improve service reliability, but it may force developers to adjust their usage patterns or accept higher bills. Additionally, DeepSeek confirmed that its legacy API endpoints — 'deepseek-chat' and 'deepseek-reasoner' — will be shut down on July 24, 2026, giving users roughly two years to migrate to the V4 models. The combination of new capabilities, peak-hour surcharges, and eventual legacy sunsetting marks a clear strategy to consolidate users onto the latest, more stable infrastructure.
- DeepSeek will launch V4-Pro and V4-Flash models in mid-July with 1M-token context windows and enhanced agentic, math, and code capabilities.
- API pricing will double during peak hours (9am-12pm and 2pm-6pm daily) to manage demand and improve stability.
- Legacy 'deepseek-chat' and 'deepseek-reasoner' APIs will be discontinued by July 24, 2026, requiring migration within two years.
Why It Matters
Developers face rising costs during peak times and a mandatory migration off legacy APIs by 2026.