From OpenRouter

Both are OpenAI-shaped, so the move is the base URL, the key, and the model ids.

- baseURL: 'https://openrouter.ai/api/v1'
+ baseURL: 'https://chat.eterial.ai/v1'

Model ids are short

OpenRouter namespaces a model by its publisher. Eterial's ids are the model, not the route to it — the same id resolves to whichever backend serves it.

OpenRouter-only fields

The request body is forwarded to the backend untouched, so these are neither stripped nor rejected — but they are not part of Eterial's contract, and on a request the GPU network serves they mean nothing at all:

• provider — routing preferences. Eterial decides the order itself; see Routing.
• models — OpenRouter's fallback array. Eterial's fallback is automatic and needs no declaring.
• transforms, route — no equivalent.
• The HTTP-Referer and X-Title headers — attribution for OpenRouter's leaderboards. Harmless, and pointless here.

One field does carry over exactly: plugins: [{ "id": "web" }] is recognised as a request for web search and routes accordingly, alongside web_search_options and a web-search entry in tools.

The bill

This is the reason to move: the same models, served by the GPU network, cost a small fraction of what an aggregator charges for them. See Pricing for the rates.

Other differences

• The catalogue is much smaller. Eterial serves a curated set rather than every model on the market. GET /v1/models is the live list.
• Capabilities narrow the routing, and a combination no single backend supports is a 400 even when /v1/models advertises both.
• Idempotency-Key is supported.