Copilot BYOK `customendpoint` rejects the model `deepseek-flash` with "502 Model not found", while `deepseek-v4-pro` works with the identical configuration. The request never reaches the DeepSeek API.

• Copilot Chat Extension Version: 1.0.84-4
• Copilot CLI: 0.66
• VS Code Version: 1.138.0
• OS Version: Windows 10
• Feature (e.g. agent/edit/ask mode): ask
• Selected model (e.g. GPT 4.1, Claude 3.7 Sonnet): deepseek-flash
• Logs:
  "Sorry, your request failed. Please try again.
  Client Request Id:
  Reason: Failed to get response from the AI model; retried 5 times (total retry wait time: 13.82 seconds) Last error: 502 Model not found"

AND
vscode output(ctrl+shift+u):
"
2026-09-20 16:11:23.133 [info] Logged in as CHICKis
2026-09-20 16:11:23.213 [info] BYOK: Copilot Chat known models list fetched successfully.
2026-09-20 16:11:23.826 [info] Got Copilot token for CHICKis
2026-09-20 16:11:23.826 [info] Copilot Chat: 0.66.0, VS Code: 1.138.0
2026-09-20 16:11:23.955 [info] copilot token sku: free_limited_copilot
2026-09-20 16:11:23.955 [info] ConversationFeature: Activating contributions
2026-09-20 16:11:23.955 [info] GitHub.vscode-pull-request-github extension is not yet activated.
2026-09-20 16:11:23.977 [info] Registering default platform agent...
2026-09-20 16:11:23.980 [info] copilot token sku: free_limited_copilot
2026-09-20 16:11:23.981 [info] [code-referencing] Public code references are enabled.
2026-09-20 16:11:23.981 [info] copilot token sku: free_limited_copilot
2026-09-20 16:11:23.981 [info] [code-referencing] Public code references are enabled.
2026-09-20 16:11:23.982 [info] copilot token sku: free_limited_copilot
2026-09-20 16:11:23.982 [info] [code-referencing] Public code references are enabled.
2026-09-20 16:11:23.982 [info] copilot token sku: free_limited_copilot
2026-09-20 16:11:23.982 [info] [code-referencing] Public code references are enabled.
2026-09-20 16:11:23.990 [info] activationBlocker from 'conversationFeature' took for 1791ms
2026-09-20 16:11:23.991 [info] [context keys] Setting context key: github.copilot-chat.activated
2026-09-20 16:11:23.996 [info] AuthenticationService: firing onDidAuthenticationChange from handleAuthChangeEvent identity change. Has token: true
2026-09-20 16:11:23.997 [info] copilot token sku: free_limited_copilot
2026-09-20 16:11:24.002 [info] GithubAvailableEmbeddingTypesManager: Got embeddings. Primary: metis-1024-I16-Binary. Deprecated:
2026-09-20 16:11:24.006 [info] WorkspaceChunkSearchService: using embedding type metis-1024-I16-Binary
2026-09-20 16:11:24.013 [error] Failed to create database. Falling back to in-memory db: Error: unable to open database file
2026-09-20 16:11:24.339 [info] Fetched model metadata in 2087ms 69894acc-df36-46e0-88c9-ec3dab172f4c
2026-09-20 16:11:24.406 [info] Latest entry: ccreq:latest.copilotmd
2026-09-20 16:11:24.408 [info] ccreq:4192350c.copilotmd | markdown
2026-09-20 16:11:24.457 [info] ConversationFeature: BYOK models available
2026-09-20 16:11:24.457 [info] copilot token sku: free_limited_copilot
"
But, direct API calls from the same machine succeed:

1. GET /models returns both deepseek-flash and deepseek-v4-pro.
2. POST /chat/completions with "model": "deepseek-flash" returns a
   valid completion.

Steps to Reproduce:

1.Custom endpoint provider with apiType: chat-completions,
URL https://api.deepseek.com. Two entries with identical settings,
only the model id differs (deepseek-flash vs deepseek-v4-pro). such as:
{ "name": "DeepSeek", "vendor": "customendpoint", "apiKey": xxxx, "apiType": "chat-completions", "models": [ { "id": "deepseek-flash", # if use "deepseek-v4-pro" is available. "name": "DeepSeek Flash", "url": "https://api.deepseek.com", "toolCalling": true, "vision": true, "maxInputTokens": 1000000, "maxOutputTokens": 384000 } ] }
2.

• Selecting "DeepSeek Flash" in the model picker fails with:
  "Sorry, your request failed... Reason: Failed to get response from the
  AI model; retried 5 times (total retry wait time: 15.93 s).
  Last error: 502 Model not found"
• Nothing is logged when the request fails — the request is rejected
  before being sent (consistent with a server-side model allowlist).
• deepseek-v4-flash (the legacy alias, still routed by DeepSeek to the
  same V4.1 Flash model) is rejected too.
• deepseek-v4-pro works fine.

Suspected cause

Copilot's customendpoint model registry/allowlist has not been updated
with the new deepseek-flash id, and the retired deepseek-v4-flash
was removed.
issue reproduces in both Copilot Chat panel and the Agents window (Ctrl+Shift+A)

No files or tests are named in the issue. Start by locating the customendpoint model registry or allowlist and reproduce the configuration with deepseek-flash, deepseek-v4-flash, and deepseek-v4-pro; trace where the request is rejected before reaching DeepSeek. Done means the current deepseek-flash identifier is accepted and the legacy alias behavior is explicitly verified.