DeepSeek legacy API model names shut off on July 24
DeepSeek says the deepseek-chat and deepseek-reasoner API aliases become inaccessible on July 24, 2026 at 15:59 UTC, so agent stacks and provider presets still using the old names need a direct V4 model ID.
DeepSeek's two legacy API model names, deepseek-chat and deepseek-reasoner, are scheduled to become inaccessible on July 24, 2026 at 15:59 UTC. The deadline matters for agent frameworks, coding assistants, proxy routers, notebooks, CI scripts, and internal presets that still pass those strings as the model value.
This is not a shutdown of DeepSeek V4. DeepSeek's own migration note says the base URL can remain unchanged, but callers should move to explicit V4 model IDs: deepseek-v4-flash or deepseek-v4-pro.
The July 24 cutoff hits the aliases, not DeepSeek V4
DeepSeek's API changelog lists two V4 models: deepseek-v4-pro and deepseek-v4-flash. It says the old names have been kept temporarily for compatibility, with deepseek-chat routing to the non-thinking mode of deepseek-v4-flash and deepseek-reasoner routing to the thinking mode of deepseek-v4-flash.
That compatibility layer is the part ending. After July 24 at 15:59 UTC, DeepSeek's release note says the two legacy names will be fully retired and inaccessible. Teams that already moved to explicit V4 IDs should not need a provider change. Teams relying on old names in wrappers may see request failures once the alias layer is removed.
Which model ID should replace each old name
For the lowest-risk migration, replace hard-coded legacy names with the model ID that matches the current compatibility behavior:
| Current model value | Current routing during transition | Safer explicit replacement |
|---|---|---|
deepseek-chat | deepseek-v4-flash non-thinking mode | deepseek-v4-flash |
deepseek-reasoner | deepseek-v4-flash thinking mode | deepseek-v4-flash with thinking mode configured |
| custom high-end reasoning preset | not guaranteed by alias mapping | evaluate deepseek-v4-pro directly |
DeepSeek describes V4 Flash as the efficient option and V4 Pro as the larger model. OpenRouter's DeepSeek V4 Flash page independently lists the V4 Flash model with a 1,048,576-token context window, reasoning support, tool-related parameters, and multiple hosted providers, which helps confirm that the migration target is available outside DeepSeek's own endpoint.
Agent stacks have more places to check than app code
The obvious fix is a one-line model change. The easy miss is where that line lives.
Check provider presets in agent tools, .env files, model-router dashboards, eval harnesses, fallback chains, prompt-debug notebooks, SDK examples copied into repos, and CI jobs that run smoke tests against DeepSeek. Also check any UI that stores a friendly name separately from the actual API slug.
Reasoning mode deserves a separate test. If a workflow used deepseek-reasoner, switching only the model string may not preserve the intended reasoning behavior unless the client or provider exposes the equivalent thinking-mode parameter. Run one short regression on tool calling, structured output, and max-token behavior rather than assuming the alias and explicit model produce identical envelopes.
Teams building or auditing agent systems can use LinkLoot's AI agent tools guide at /guides/ai-agent-tools as a checklist companion, but the urgent task is narrower: find the deprecated strings and remove them.
Source check
The primary evidence is DeepSeek's API changelog, which names the July 24, 2026 15:59 UTC cutoff and the V4 replacement IDs. OpenRouter corroborates public availability of deepseek/deepseek-v4-flash, including model metadata and provider availability. WaveSpeedAI published a migration guide that matches the deadline and replacement path, but DeepSeek's own docs remain the source of record for the cutoff.
If DeepSeek extends the compatibility window, the official API changelog should be the first page to change. Until then, treat July 24 at 15:59 UTC as the operational deadline.