DeepSeek's official docs now publish a GitHub Copilot CLI BYOK route with Anthropic endpoint wiring, a 400-error avoidance rule, and explicit token limits

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 official-source monitoring note

Today's accepted item stays DeepSeek-first and uses a current official DeepSeek documentation page because the public X surface still does not expose a newer safely verifiable @deepseek_ai post than the older April pricing thread, which is both stale and already overlapped by existing pricing coverage on this site.

What we verified on June 16, 2026

• DeepSeek now has an official GitHub Copilot CLI integration page in its agent-integration docs.
• That page explicitly tells users to use anthropic as the provider type when wiring GitHub Copilot CLI to DeepSeek via BYOK.
• The same page warns that the openai provider path triggers a 400 error because Copilot CLI's OpenAI integration does not echo reasoning_content back on later requests, while the Anthropic Messages route avoids that failure mode.
• DeepSeek's documented environment contract is concrete: COPILOT_PROVIDER_BASE_URL=https://api.deepseek.com/anthropic, a DeepSeek API key, and COPILOT_MODEL=deepseek-v4-pro, with deepseek-v4-flash also listed as an available model.
• The same page adds explicit token-limit guidance for custom DeepSeek model wiring inside Copilot CLI: 840000 prompt tokens and 128000 output tokens.
• DeepSeek also documents an optional offline mode that blocks GitHub API calls while prompts still go to api.deepseek.com.

Why this is publishable

This is not framed as a new stocked product or a speculative roadmap item. It is a current official developer-operations signal that changes how people should configure a real coding-agent workflow:

1. It gives a vendor-published answer to a concrete integration failure instead of leaving developers to debug a generic 400 on their own.
2. It adds a second Copilot-adjacent official DeepSeek surface after the earlier VS Code Copilot Chat guide, but with a different endpoint contract and different operational constraints.
3. It is not a duplicate of the June 14 GitHub Copilot Chat item, the June 15 Hermes guide, or the June 10 Claude Code page.

Why this matters for DeepSeek-first SEO pages

• GitHub Copilot CLI setup pages can now explain why Anthropic-format routing is the official DeepSeek path, not just list environment variables blindly.
• API-routing guides should separate Copilot Chat extension behavior from Copilot CLI BYOK behavior, because the former uses an extension flow while the latter depends on environment-based provider wiring.
• Cost-aware coding-agent pages should mention that custom token ceilings are part of the official DeepSeek setup contract when Copilot CLI does not already know the model's limits.

Rejected candidates today

• The latest publicly visible @deepseek_ai X result about the April 26 input-cache price drop: official, but older than the current docs-backed developer signals and too close to the site's existing pricing and price-cut coverage.
• The same GitHub Copilot Chat page from June 14: still official, but already published and therefore a duplicate-content risk.
• The Hermes integration page from June 15: still official, but already published yesterday.
• Status page uptime alone: official, but weaker and less actionable than a docs page that changes real CLI setup behavior.

Editorial takeaway

The safest official DeepSeek story today is a GitHub Copilot CLI setup contract: DeepSeek's own docs now tell teams to use the Anthropic-compatible endpoint, explain the 400 error that appears on the OpenAI path, and publish explicit token-limit values for V4 Pro. That is stronger and more useful than recycling an older social post.

Sources checked

• DeepSeek English homepage
• DeepSeek official GitHub Copilot CLI integration guide
• DeepSeek official GitHub Copilot integration guide
• DeepSeek official status page
• DeepSeek official X account surface