Model alias is case-sensitive and causes 400 for lowercase Kimi model name

[andrey055]

Summary

OpenGNK forwards the model field to the upstream Gonka API without normalizing known model aliases.

As a result, requests using the lowercase model name:

{
  "model": "moonshotai/kimi-k2.6"
}

fail with HTTP 400, even though the intended model exists as:

{
  "model": "moonshotai/Kimi-K2.6"
}

This is inconvenient for OpenAI-compatible clients or gateways that may normalize model identifiers to lowercase.

Environment

• Repository: gonkalabs/opengnk
• Endpoint: /v1/chat/completions
• Mode: streaming request
• Upstream: https://node4.gonka.ai

Error

time=2026-05-14T06:28:52.307Z level=INFO msg="upstream stream request" method=POST url=https://node4.gonka.ai/v1/chat/completions endpoint_addr=gonka1y2a9p56kv044327uycmqdexl7zs82fs5ryv5le wallet=gonka1m5ncrzrjks9utrz98fjdlak8n6gslnz30rfu62

time=2026-05-14T06:28:52.418Z level=ERROR msg="upstream stream status" code=400 body="{\"error\":{\"message\":\"unsupported model \\\"moonshotai/kimi-k2.6\\\"; supported models: moonshotai/Kimi-K2.6, Qwen/Qwen3-235B-A22B-Instruct-2507-FP8\"}}\n"

Steps to reproduce

Send a request with lowercase Kimi model name:

curl -s http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "moonshotai/kimi-k2.6",
    "messages": [
      {
        "role": "user",
        "content": "hi"
      }
    ],
    "max_tokens": 20,
    "stream": true
  }'

Actual behavior

OpenGNK forwards the model name as-is to the upstream API.

The upstream API returns:

{
  "error": {
    "message": "unsupported model \"moonshotai/kimi-k2.6\"; supported models: moonshotai/Kimi-K2.6, Qwen/Qwen3-235B-A22B-Instruct-2507-FP8"
  }
}

Expected behavior

OpenGNK should normalize known model aliases before forwarding the request upstream.

For example:

moonshotai/kimi-k2.6 -> moonshotai/Kimi-K2.6
qwen/qwen3-235b-a22b-instruct-2507-fp8 -> Qwen/Qwen3-235B-A22B-Instruct-2507-FP8

Suggested fix

Add model alias normalization in internal/api/handler.go, inside chatCompletions, after reading/sanitizing the request body and before forwarding it to streamResponse, nonStreamResponse, or tool simulation logic.

This would make the proxy more compatible with OpenAI-style clients and gateways that may lowercase model identifiers.

Impact

• Some OpenAI-compatible clients fail when using lowercase model identifiers.
• Users receive a 400 error even though the model exists and is available.
• The issue affects compatibility rather than core inference functionality.