If you or your team is running production workloads on DeepSeek’s API, mark July 24, 2026 on the calendar: that’s when the legacy deepseek-chat and deepseek-reasoner model aliases stop responding entirely, with no announced extension. DeepSeek is forcing migration to its V4 family as it moves that generation from preview to official stable release — and unlike most model deprecations, this one comes with a hard cutoff rather than a gradual sunset.
What’s changing
DeepSeek V4 graduates from preview to stable release around the same mid-July window, and the older aliases are being retired outright rather than left running in parallel. Any application, script, or workflow still calling deepseek-chat or deepseek-reasoner directly will start returning errors after the cutoff. There’s no grace period built in based on current announcements.
What to do before July 24
Audit any codebase, automation, or third-party tool integration that references the old model aliases and update them to the corresponding V4 model IDs before the deadline. If you’re using DeepSeek through an aggregator or gateway (OpenRouter, a unified API service, etc.) rather than DeepSeek’s API directly, check whether that layer has already handled the routing for you — some have, some haven’t, and it’s worth confirming rather than assuming.
Context: why this matters beyond DeepSeek users
DeepSeek remains one of the most widely used open-weight model families for cost-sensitive production workloads, and this is a reminder that even open-weight providers can force hard migrations on their hosted API — self-hosting the older weights is still an option for anyone who wants to stay on the old version rather than upgrade. It’s also a useful case study in how fast the frontier is moving: DeepSeek is retiring an API generation in the same month that GPT-5.6, Gemini 3.5 Pro, and Grok 4.5 are all converging on new releases.
Last updated: July 17, 2026.