Raptor Mini returns "Internal Server Error" but is available in VSCode

[hatlesswizard]

Hello!

The problem: When I use oswe-vscode-prime mode, I receive "Internal Server Error".

What have I done:

1. Tried other models like gpt-5-mini, they work.
2. Tried logging out and logging in.
3. Tried logging in with completely different account.

The error I receive:

➜ Listening on: http://localhost:4141/ (all interfaces)
<-- POST /v1/chat/completions
ℹ Current token count: { input: 8, output: 0 }                                                                                                                 10:35:08 AM

 ERROR  Failed to create chat completions Response { status: 500,                                                                                               10:35:16 AM
  statusText: 'Internal Server Error',
  headers:
   Headers { 'content-security-policy': "default-src 'none'; sandbox",
     'content-type': 'text/plain; charset=utf-8',
     'copilot-edits-session': '1770964509.e3cb2dfa61360e61bcbf5f02f74d092f561569c04b6259f42c2175d1cc4a7140',
     'strict-transport-security': 'max-age=31536000',
     'x-content-type-options': 'nosniff',
     'x-request-id': '771c8775-1dd2-4b96-817c-2a6b4564062a',
     date: 'Fri, 13 Feb 2026 06:35:16 GMT',
     'content-length': '22',
     'x-github-backend': 'Kubernetes',
     'x-github-request-id': 'B7FB:38E450:27019F:2E0D6F:698EC61D' },
  body: ReadableStream { locked: false, state: 'readable', supportsBYOB: true },
  bodyUsed: false,
  ok: false,
  redirected: false,
  type: 'basic',
  url: 'https://api.githubcopilot.com/chat/completions' }


 ERROR  Error occurred: Failed to create chat completions                                                                                                       10:35:16 AM

    at createChatCompletions (/Users/desktop-lthlb72/.npm/_npx/b3c0db1118356182/node_modules/copilot-api/dist/main.js:783:9)
    at process.processTicksAndRejections (node:internal/process/task_queues:103:5)
    at async handleCompletion$1 (/Users/desktop-lthlb72/.npm/_npx/b3c0db1118356182/node_modules/copilot-api/dist/main.js:812:19)
    at async /Users/desktop-lthlb72/.npm/_npx/b3c0db1118356182/node_modules/copilot-api/dist/main.js:832:10
    at async dispatch (/Users/desktop-lthlb72/.npm/_npx/b3c0db1118356182/node_modules/hono/dist/compose.js:22:17)
    at async cors2 (/Users/desktop-lthlb72/.npm/_npx/b3c0db1118356182/node_modules/hono/dist/middleware/cors/index.js:79:5)
    at async dispatch (/Users/desktop-lthlb72/.npm/_npx/b3c0db1118356182/node_modules/hono/dist/compose.js:22:17)
    at async logger2 (/Users/desktop-lthlb72/.npm/_npx/b3c0db1118356182/node_modules/hono/dist/middleware/logger/index.js:38:5)
    at async dispatch (/Users/desktop-lthlb72/.npm/_npx/b3c0db1118356182/node_modules/hono/dist/compose.js:22:17)
    at async /Users/desktop-lthlb72/.npm/_npx/b3c0db1118356182/node_modules/hono/dist/hono-base.js:301:25


 ERROR  HTTP error: Internal Server Error                                                                                                                       10:35:16 AM


--> POST /v1/chat/completions 500 7s

The command I send:

curl http://localhost:4141/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "oswe-vscode-prime",
    "messages": [
      { "role": "user", "content": "hello" }
    ]
  }'

What I personally think: The 500 error seems to come from the Copilot backend rather than my local setup. Since the model works inside VS Code, it’s possible that recent backend or API changes are causing this issue.

Could you please advise on a possible solution/workaround?

[dzeveckij]

Copilot team does not want Raptor mini to be used outside of VS Code, hence they started restricting it when a request is made via a custom API call. Honestly, I wouldn’t risk it, they could ban you for it.

[hatlesswizard]

The model is way too good and speed is great. I am willing to risk it if there is a workaround.

[jeiksegovia]

Copilot team does not want Raptor mini to be used outside of VS Code, hence they started restricting it when a request is made via a custom API call. Honestly, I wouldn’t risk it, they could ban you for it.

Model is very good, I find gpt-5-mini "Balanced" but for coding tasks, oswe-prime-vscode is superior.
It's worths the shot

[dzeveckij]

The model is way too good and speed is great. I am willing to risk it if there is a workaround.

Is the speed actually great? I remember on first release I tried it and it was extremely slow compared to Grok Code Fast and even slower than gpt-5-mini, so I didn’t use it much after that.

Model is very good, I find gpt-5-mini "Balanced" but for coding tasks, oswe-prime-vscode is superior. It's worths the shot

I only tried it on release, felt similar to gpt-5-mini, just spent more time reasoning. I don’t see much discussion since not many people use it outside Copilot. Were there any significant improvements after release?

[jeiksegovia]

The model is way too good and speed is great. I am willing to risk it if there is a workaround.

Is the speed actually great? I remember on first release I tried it and it was extremely slow compared to Grok Code Fast and even slower than gpt-5-mini, so I didn’t use it much after that.

Model is very good, I find gpt-5-mini "Balanced" but for coding tasks, oswe-prime-vscode is superior. It's worths the shot

I only tried it on release, felt similar to gpt-5-mini, just spent more time reasoning. I don’t see much discussion since not many people use it outside Copilot. Were there any significant improvements after release?

Yup it's the same model, but optimize for code agentic task in my opinion. I use it as FAST model for claude-code: with a bigger context window (raptor variant) and more literal when following instructions it's perfect for tasks delegated from Main model like: scan web pages snapshots, scanning logs files and extract stuff, sanity checks "is this bash command secure/compliant with this set of rules". It works very good as a cheap efficient sub-sub-agent tool.