DeepSeek's Claude Code docs now need stricter model pins as old chat and reasoner aliases head toward July 24 deprecation

Today's accepted developer signal is on the Claude Code track: DeepSeek's official docs now tie Claude-style integrations more clearly to exact V4 model strings, while the main API quick-start warns that `deepseek-chat` and `deepseek-reasoner` will be deprecated on July 24, 2026.

英文原文

这篇动态暂时没有完整中文稿。为了避免中文 SEO 页面收录英文正文，详情页会标记为 noindex。

Accepted developer update

Today's safe publishable change belongs on the DeepSeek in Claude Code track, not on pricing or broader model-market coverage.

DeepSeek's official Claude Code page still gives the same Anthropic-compatible baseline that this site already recommends, but the broader official API docs now add an important migration warning that older community tutorials can easily miss: deepseek-chat and deepseek-reasoner are scheduled to be deprecated on July 24, 2026.

What changed upstream

The official Claude Code recipe still expects exact V4 strings: ANTHROPIC_MODEL=deepseek-v4-pro[1m], the matching ANTHROPIC_DEFAULT_* pins, CLAUDE_CODE_SUBAGENT_MODEL=deepseek-v4-flash, and CLAUDE_CODE_EFFORT_LEVEL=max.
The official API quick-start now flags legacy aliases with a real date: deepseek-chat and deepseek-reasoner are marked for 2026-07-24 deprecation.
The compatibility mapping is now explicit: the same quick-start says those old names only exist for compatibility and currently map to the non-thinking and thinking modes of deepseek-v4-flash, respectively.
The Anthropic-format warning still matters: DeepSeek's Anthropic API docs say unsupported model names can automatically map to deepseek-v4-flash. That means outdated aliases or altered provider-side names can silently leave a Claude Code or Desktop workflow on Flash when the user expected Pro.

Why this matters for crawlable Claude Code setup pages

This is not just a naming nit. It changes what a trustworthy DeepSeek-via-Claude setup guide should tell developers to do right now:

Copy the official V4 model strings exactly instead of reusing older deepseek-chat or deepseek-reasoner snippets.
Treat July 24, 2026 as the migration deadline for any Claude Code, Claude Desktop, Gateway, or settings.json workflow still using those aliases.
Verify provider-side model usage when testing Desktop mapping or "Cloud Code" wording variants, because unsupported names can still fall back to Flash.

What we rejected today

DeepSeek TUI: official repo, release, and developer-site activity were rechecked, but there was no newer stable release or merged install/runtime doc change beyond the May 19 publish already live on this site.
DeepSeek local development: the official DeepSeek V4 Flash model card still points to the same vLLM, SGLang, Docker Model Runner, and curl validation baseline already covered in the maintained local-deployment guide.
Community Desktop / gateway / local-runtime chatter: useful for discovery, but not strong enough to publish without a clearer official or reproducible delta.

Editorial takeaway

Today's public update belongs on the Claude Code guide, the dedicated /claude-code landing page, and the news stream. It does not affect stocked plans, plan cards, or any purchasable inventory surface.