Reference

FAQ

Find concise answers to common CLSSAI integration and operations questions.

On this page

This page is for evaluators and developers who need concise answers before or during an integration.

What endpoints are available?#

Purpose	Method and endpoint
OpenAI-compatible model list (public)	`GET https://api.clssai.com/v1/models`
Chat Completions	`POST https://api.clssai.com/v1/chat/completions`
Responses API (verified; data-only SSE, dispatch on JSON `type`)	`POST https://api.clssai.com/v1/responses`
Legacy Completions (text models; wrong-modality models are rejected)	`POST https://api.clssai.com/v1/completions`
Embeddings	`POST https://api.clssai.com/v1/embeddings`
Reranking	`POST https://api.clssai.com/v1/rerank`
Image generation	`POST https://api.clssai.com/v1/images`
Image edits (multipart translated to JSON; no mask; 8 MB total; always `b64_json`)	`POST https://api.clssai.com/v1/images/edits`
Speech synthesis	`POST https://api.clssai.com/v1/audio/speech`
Transcription	`POST https://api.clssai.com/v1/audio/transcriptions`
Video pass-through (lightly tested)	`POST https://api.clssai.com/v1/videos`
Key balance	`GET https://api.clssai.com/v1/balance`
Anthropic Messages	`POST https://api.clssai.com/v1/messages`
Anthropic token count (not billed)	`POST https://api.clssai.com/v1/messages/count_tokens`
Anthropic model list (public)	`GET https://api.clssai.com/anthropic/v1/models`
Gemini generation	`POST https://api.clssai.com/v1beta/models/{model}:generateContent`
Gemini streaming	`POST https://api.clssai.com/v1beta/models/{model}:streamGenerateContent`
Gemini model list (public)	`GET https://api.clssai.com/v1beta/models`
Request logs	`GET https://api.clssai.com/key/<key>/logs.json`

OpenAI-compatible calls use the base URL https://api.clssai.com/v1 and accept Bearer, x-api-key, or a ?key= query parameter. Prefer a header: a key in the URL leaks through referrers, proxy logs, and browser history. The Anthropic SDK base URL is https://api.clssai.com without /v1; calls accept x-api-key or Bearer and require anthropic-version: 2023-06-01. Gemini uses https://api.clssai.com/v1beta with x-goog-api-key or ?key= and does not accept Bearer authentication.

Which models can I call?#

Use the public GET /v1/models response as the authority for callable OpenAI-compatible IDs. OpenAI-compatible routes accept full IDs and exact bare aliases, but production clients should prefer full IDs and confirm resolution in the response model field. Anthropic and Gemini use their native names from GET /anthropic/v1/models and GET /v1beta/models. The Models & pricing page documents the rules, while /models provides the full OpenAI-compatible browser.

How is a request priced?#

An upstream-reported usage.cost is authoritative when present; otherwise billing uses the local price table. Reference catalog prices are estimates for selection, not a final invoice. See Usage & billing.

When does my balance update?#

Settlement runs through an asynchronous queue and can lag by minutes. A completed response and the displayed aggregate balance are not atomic. Request logs are the best unit for reconciliation.

Is there free access?#

Some complete model IDs end in :free. Free users also share a separate global daily pool; when it is exhausted, the gateway returns 429. A small grace amount exists but is not a published balance.

What data-policy commitment applies?#

The gateway capability inventory does not establish a universal training or retention promise for every upstream model. Review your own data requirements and contact Support before sending sensitive workloads.

Can I fine-tune a model here?#

This documentation establishes inference endpoints only and does not define a fine-tuning contract. Use only the endpoints documented in this site.

Can I deploy CLSSAI privately?#

This documentation does not define a private-deployment offering or contract. Contact Support for current commercial requirements.

What are the rate limits?#

CLSSAI does not add a per-key RPM or concurrency limit. A gateway 429 can mean the global free daily pool is exhausted; other 429 responses come from the upstream API. See Errors & rate limits.

Can I call CLSSAI from browser JavaScript?#

No. Some inference responses return Access-Control-Allow-Origin: *, and a preflight-free request (for example a ?key= URL with Content-Type: text/plain) can complete from a page, so a browser call may appear to work. It still ships your API key to every visitor. Call CLSSAI from your backend and expose only the minimum safe functionality to your frontend. See Authentication & keys.

How do I report a failed inference request?#

Send Support the timestamp, endpoint, HTTP status, model ID, and cf-ray response header. For OpenAI-compatible routes, include x-generation-id when present. Never send the complete API key.

Common mistakes#

Include /v1 in OpenAI SDK base URLs and omit it from the Anthropic SDK base URL.
Keep inference calls and keys on a server.
Preserve the full author/model:free or author/model:batch ID.
Check an empty choices array in streaming usage frames.
Use cf-ray to trace inference requests.

Need a hand?

Find answers to common questions or diagnose a failed request.

Frequently asked questions →Troubleshoot errors →