OpenAI-compatible API providers: what the label really means
"OpenAI-compatible" is the least standardised phrase in the AI stack. Two providers can both use it and behave completely differently the moment you hit a stream, a tool call, or a rate limit. Here is what to check before you migrate.
Four things that break in migration
- Streaming chunk shape. Most providers copy the SSE format, but usage numbers (prompt/completion tokens) often arrive in the final chunk - or not at all. If you rely on usage for billing, verify it.
- Tool / function calling. Supported by many, but argument coercion and parallel-call behaviour differ. Test with 20 real calls, not one.
- Error taxonomy. A 400 for "context too long" vs a 500 tells your retry logic different things. Ask for the error list on day one.
- Model ID naming. The OpenAI SDK does not validate model names, so a typo becomes a runtime error in production.
How we handle these
Same request shape as OpenAI, streaming SSE with a final usage chunk, explicit error codes for context length / model unavailable / insufficient balance, and a public model list you can diff in CI.
Minimal migration
from openai import OpenAI
client = OpenAI(
base_url="https://tapixai.com/v1",
api_key="sk-fg-..."
)
r = client.chat.completions.create(
model="gemini-3-flash",
messages=[{"role": "user", "content": "hello"}]
)
Everything else in your codebase stays the same: same SDK, same retries, same streaming iterator.
Live prices from our catalogue
Read from our live catalogue on every page load, so these numbers cannot go stale. USD per 1M tokens.
| Model | Input / 1M | Output / 1M |
|---|---|---|
| — | — | — |
How to compare two "compatible" providers in 30 minutes
- Send 10 prompts at temperature 0 on the same model, diff the outputs and the latency.
- Kill the connection mid-stream and see whether the provider bills you for the partial answer.
- Send an oversized prompt and check the status code you get back.
- Compare the price per 1M output tokens, not input - output is where the bill lives.