Developers relying on DeepSeek’s legacy API aliases face a hard deadline: on July 24, 2026, at 15:59 UTC, the company will permanently retire deepseek-chat and deepseek-reasoner. This transition forces a mandatory migration to the V4 model family. Because the V4 suite remains in technical preview with no announced general availability date, engineering teams are currently anchoring production agentic workloads to software that the provider itself classifies as experimental.
The migration path is technically minimal. DeepSeek has maintained the existing base URL, requiring only a swap of the model string to either deepseek-v4-pro or deepseek-v4-flash. V4-Flash is positioned as a high-volume workhorse, priced at $0.14 per million input tokens and $0.28 per million output tokens. This creates a stark cost delta: approximately 36 times cheaper on input and over 100 times cheaper on output than GPT-5.5 or Claude Opus 4.7.
DeepSeek has introduced peak-valley pricing, where rates during peak hours — 09:00-12:00 and 14:00-18:00 Beijing Time — are approximately double the off-peak rate. This adds complexity to operational budgeting.
Hardware dependency defines the underlying risk. DeepSeek V4 is natively optimized for the Huawei Ascend 950 processor. For US-based enterprise teams, this creates a tangible procurement and export-control risk. Many organizations are hedging by maintaining parallel accounts with Western providers.