Updated 2026-06-26
DeepSeek chat and reasoner retirement date: migrate before July 24, 2026 instead of treating old aliases as permanent
DeepSeek's current change log and V4 release docs are explicit: `deepseek-chat` and `deepseek-reasoner` are compatibility aliases, not stable forever model names. They are scheduled to be discontinued on July 24, 2026, and during the transition both aliases point to Flash-family behavior rather than preserving an independent long-term route. That makes retirement-date searches a real support topic, not a documentation footnote.
1. What DeepSeek officially says will happen on July 24, 2026
DeepSeek's official change log says the API now supports `deepseek-v4-pro` and `deepseek-v4-flash`, and that the legacy names `deepseek-chat` and `deepseek-reasoner` will be discontinued on July 24, 2026.
The same official note explains the current mapping rule: `deepseek-chat` points to the non-thinking mode of V4 Flash, while `deepseek-reasoner` points to the thinking mode of V4 Flash. That means an app still using those aliases is already on a compatibility bridge rather than on a separate long-term model line.
Sources checked
- DeepSeek official change log - Primary source for the July 24, 2026 discontinuation date and current alias mapping.
- DeepSeek V4 Preview release note - Reinforces that V4 Pro and V4 Flash are the current official model names.
2. Translate old model names into the current DeepSeek routing logic
A practical migration starts by separating old labels from current behavior. `deepseek-chat` is now the non-thinking Flash route. `deepseek-reasoner` is the thinking Flash route. Neither alias gives you a durable guarantee after July 24, 2026.
If your workload only needed a cheap default text route, move it to `deepseek-v4-flash` directly. If your app depended on heavier reasoning, architecture review, or quality-sensitive coding, test `deepseek-v4-pro` as the explicit target instead of assuming `deepseek-reasoner` should remain the premium route.
| Legacy model name | Current official behavior | Preferred explicit replacement |
|---|---|---|
| `deepseek-chat` | V4 Flash non-thinking mode | `deepseek-v4-flash` |
| `deepseek-reasoner` | V4 Flash thinking mode | `deepseek-v4-flash` or `deepseek-v4-pro` after testing |
3. Update OpenAI-format clients without rebuilding the rest of the stack
DeepSeek's V4 release note says the base URL remains unchanged. For most OpenAI-compatible clients, the migration is mostly a model-name change rather than a transport rewrite.
That is useful because many codebases can keep the same SDK, key-loading pattern, retries, and streaming logic. The cleanup work is mainly around model IDs, routing policy, and tests that still assert old alias names.
from openai import OpenAI
client = OpenAI(
api_key="sk-...",
base_url="https://api.deepseek.com",
)
response = client.chat.completions.create(
model="deepseek-v4-flash",
messages=[{"role": "user", "content": "Summarize this changelog in JSON."}],
)Sources checked
- DeepSeek official API quick-start - Shows the unchanged base URL and current OpenAI-compatible positioning.
4. Update Anthropic-format clients and Claude-style tooling carefully
If your stack uses the Anthropic-compatible route, keep the migration separate from the old alias cleanup. DeepSeek's Anthropic API guide uses `https://api.deepseek.com/anthropic` and explicit V4 model names.
That same guide warns that unsupported model names can automatically map to `deepseek-v4-flash`. So a sloppy migration can appear successful while silently routing to Flash. Verify provider-side usage instead of trusting a client label alone.
export ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic
export ANTHROPIC_API_KEY=sk-...
python - <<'PY'
import anthropic
client = anthropic.Anthropic()
message = client.messages.create(
model="deepseek-v4-pro",
max_tokens=256,
messages=[{"role": "user", "content": [{"type": "text", "text": "Ping"}]}],
)
print(message.content)
PYSources checked
- DeepSeek official Anthropic API guide - Primary source for the Anthropic base URL and unsupported-model fallback warning.
5. Use a migration checklist instead of changing names blindly
Search the repo for old alias names in application code, environment files, CI variables, dashboards, and fallback configs. Alias strings often survive in more than one place.
Then validate the behavioral boundary: which routes should stay on Flash, which requests deserve Pro, and which prompts need thinking mode explicitly enabled. If you need the parameter-level thinking rules next, continue with `/guides/deepseek-thinking-mode-parameter-rules`.
For broader alias cleanup context, compare `/guides/deepseek-chat-to-v4-migration` and `/guides/deepseek-anthropic-model-fallback` after this page.
rg -n "deepseek-chat|deepseek-reasoner" .
# then update model IDs intentionally instead of doing a blind global replace6. Where this guide fits on a DeepSeek-first site
This page is migration support content. It does not create a new product, and it does not mean every model shown in docs is sold on `/pricing`.
The correct user path is narrow: clean up deprecated aliases here, then use `/pricing` only if you need an in-stock DeepSeek API key route that the site actually sells.
FAQ
When do `deepseek-chat` and `deepseek-reasoner` retire?
DeepSeek's official change log says both legacy names will be discontinued on July 24, 2026.
What does `deepseek-chat` point to right now?
DeepSeek says `deepseek-chat` currently points to the non-thinking mode of `deepseek-v4-flash` during the transition period.
What does `deepseek-reasoner` point to right now?
DeepSeek says `deepseek-reasoner` currently points to the thinking mode of `deepseek-v4-flash` during the transition period.
Should every old reasoner workflow move to V4 Pro?
Not automatically. Some traffic can stay on V4 Flash with thinking enabled, but quality-sensitive work should be tested against V4 Pro explicitly.
Does this page mean V4 Pro or Flash are guaranteed purchasable plans here?
No. Guide content explains routing and migration. Purchasable cards still depend on actual stocked inventory on `/pricing`.
The practical alias-cleanup rule is simple: treat July 24, 2026 as a real retirement date, stop relying on `deepseek-chat` and `deepseek-reasoner` as durable names, and migrate to explicit V4 model IDs with verification instead of trusting compatibility aliases to keep behaving forever.
Related model comparisons
Continue from this guide into structured DeepSeek-first comparison pages with model tables, routing advice, and pricing context.