Skip to content
NewsProduct launch

DeepSeek retires legacy API aliases July 24: four days to rename model strings or face 4xx errors

· by Pondero Newsdesk

The short version

DeepSeek sunsets deepseek-chat and deepseek-reasoner on July 24, 2026 at 15:59 UTC. Production systems still using the legacy names have four days to swap in deepseek-v4-flash or deepseek-v4-pro before calls start failing.

DeepSeek retires legacy API aliases July 24: four days to rename model strings or face 4xx errors

Production API calls using deepseek-chat or deepseek-reasoner will start returning errors in four days. Per DeepSeek's official API docs, both aliases become fully inaccessible after July 24, 2026 at 15:59 UTC. The fix is a one-line string replacement.

What is changing

DeepSeek launched V4 on April 24, 2026, splitting the model into two variants. V4-Pro carries 1.6 trillion total parameters with 49 billion active per token using a mixture-of-experts architecture. V4-Flash runs at 284 billion total parameters with 13 billion active, built for speed and cost efficiency. Both support a 1-million-token context window, per the release announcement.

Since April 24, the legacy aliases have routed silently: deepseek-chat maps to V4-Flash in non-thinking mode, and deepseek-reasoner maps to V4-Flash in thinking mode. That silent routing ends on July 24 at 15:59 UTC. After that cutoff, requests on either old name will fail outright.

The migration requires changing one string in each API call. The base URL (https://api.deepseek.com), the API key, and every other request parameter stay the same. Replace deepseek-chat with deepseek-v4-flash for a like-for-like swap on cost and speed tier, or with deepseek-v4-pro for the full-capability model. The same swap applies to LangChain, CrewAI, AutoGen, and any other framework that passes a model name string to the OpenAI-compatible endpoint.

Why it matters

The deadline is firm. DeepSeek extended the V4 preview period twice before setting July 24 as the GA retirement date, but the official notice gives no indication of another extension. Any application that misses the window returns 4xx errors to end users with no automatic fallback.

The rename also forces a real decision. V4-Pro and V4-Flash are not the same model at different prices. Callers using deepseek-chat for heavyweight reasoning, multi-step agent tasks, or agentic coding workflows may find the straight swap to V4-Flash degrades output quality. Those tasks are better served by V4-Pro. Simple retrieval, classification, and high-volume calls at low latency fit V4-Flash. The published pricing as of July 2026: V4-Flash at $0.14 per million cache-miss input tokens and $0.28 per million output tokens; V4-Pro at $0.435 per million input tokens and $0.87 per million output tokens.

Running test calls on both model strings before Thursday is the fastest way to confirm behavior before the cutoff forces the issue.

What to watch next

DeepSeek has not issued any deadline extension notice. The API docs news page is the canonical source for any last-minute update. Developers using DeepSeek V4 via third-party inference providers such as Fireworks AI or Together AI should check those platforms' documentation separately; those providers may use different model string conventions that do not map one-to-one to DeepSeek's own aliases.

Sources