DEV Community

Natalie Yevtushyna
Natalie Yevtushyna

Posted on • Originally published at gonkarouter.io

OpenAI-Compatible" Means Format Compatible, Not Official Model Access

#ai #api #opensource #llm

Every AI gateway's landing page says "OpenAI-compatible" somewhere near the top. It's become close to meaningless as a marketing phrase, because it's true of almost everything in this category, and it's also frequently misread as something it isn't.

What "compatible" actually means

It means: the request and response shape matches OpenAI's (or Anthropic's) API. Same JSON structure, same auth header pattern, same client library works with a base_url swap. It does not mean:

  • You're getting OpenAI's or Anthropic's actual models
  • Behavior is identical to calling OpenAI or Anthropic directly
  • Every parameter, every edge case, every streaming quirk is preserved

GonkaRouter is refreshingly direct about this distinction in its own materials, worth quoting because it's the right way to phrase it: "GonkaRouter's OpenAI-compatible API format does not mean official OpenAI model access. Its Anthropic-compatible API access does not mean official Anthropic model access." That's the sentence every gateway should have somewhere, and a lot of them don't say it this plainly.

Why this distinction is the thing to actually check

If your integration testing only confirms "the client library didn't throw an error," you've verified format compatibility, not behavioral compatibility. The gap between those two shows up in exactly the places that are annoying to debug after the fact:

  • Streaming behavior (does it chunk the same way?)
  • Tool/function calling schema strictness
  • How system prompts get weighted or truncated
  • Error format on rate limits or invalid requests
  • Context window handling at the edges

None of this means "compatible" gateways are bad, it means the compatibility claim is about the interface, and you still need to test the actual model behavior against your specific workload before trusting it in production.

Currently supported models (worth checking against your own needs, not this post)

GonkaRouter's current lineup, per their own docs: MiniMax-M2.7, Kimi-K2.6, and GLM-5.2. Model lists in this space move fast and I've seen slightly different lineups mentioned across their own posts (some mention Qwen instead of GLM), so treat any specific model list, including this one, as a snapshot, not a guarantee, check gonkarouter.io/models directly before you build against a specific model ID.

from openai import OpenAI

client = OpenAI(
    api_key="sk-xxxxxx",
    base_url="https://api.gonkarouter.io/v1",
)

response = client.chat.completions.create(
    model="MiniMaxAI/MiniMax-M2.7",
    messages=[{"role": "user", "content": "Hello!"}],
)
Enter fullscreen mode Exit fullscreen mode

The honest evaluation checklist

  • [ ] Confirmed the format matches (client library works without errors)
  • [ ] Tested actual model behavior against your real prompts, not just a hello-world call
  • [ ] Checked streaming behavior specifically if your app depends on it
  • [ ] Verified error format for the failure modes you actually care about (rate limits, invalid input)
  • [ ] Didn't assume "compatible" means "identical to the provider it's compatible with"

Has anyone run into a compatibility gap that only showed up after real production traffic, not initial testing? Curious what category it fell into, streaming, tool calling, or something else entirely.

Top comments (0)