Unsupported endpoints and features
The gateway accepts a path only when it can meter its cost exactly. Paths that cannot be metered (such as OpenAI Assistants, a file with a purpose other than batch, or Gemini cachedContents) are refused with 402 and code path_unmetered, before anything is sent or billed. This page is the full list, so that you know before you build.
Last updated:
On the OpenAI routes
| What | Path | Answer |
|---|---|---|
Files with a purpose other than batch, Uploads | Path:/v1/files, /v1/uploads | Answer:402 path_unmetered |
| Other Files and Batches paths | Path:Under /v1/files and /v1/batches, any path but the batch routes | Answer:402 path_unmetered |
| Assistants, Threads | Path:/v1/assistants, /v1/threads | Answer:402 path_unmetered |
| Vector stores, Containers | Path:/v1/vector_stores, /v1/containers | Answer:402 path_unmetered |
| Conversations | Path:/v1/conversations | Answer:402 path_unmetered |
| Fine-tuning, Evals | Path:/v1/fine_tuning, /v1/evals | Answer:402 path_unmetered |
| Stored responses and chat completions | Path:/v1/responses/{id}, /v1/chat/completions/{id} | Answer:402 path_unmetered |
| The organization admin API | Path:/v1/organization | Answer:402 path_unmetered |
| OpenAI's own Realtime client secrets | Path:/v1/realtime/client_secrets | Answer:402 path_unmetered; use /v1/realtime/client_secret (realtime voice) |
| Background execution on Responses | Path:background: true | Answer:400 path_unmetered |
| Hosted tools | Path:code_interpreter, file_search, image_generation, computer_use, mcp | Answer:400 hosted_tool_unmetered |
| Streamed image generation | Path:stream: true | Answer:400 path_unmetered |
| Service tiers | Path:service_tier | Answer:400 invalid_request |
Because store is always false, features that rely on a response stored at OpenAI are not available. The Batch API and the upload of its input files (purpose=batch) are supported, at 50% of the usual price: OpenAI Batch.
On the Anthropic route
- Under
/anthropiconlyPOST /v1/messages,POST /v1/messages/count_tokens,GET /v1/modelsand Message Batches (without the batch list) are documented and supported. Other paths (such as the Files API and the Admin API) answer402with codepath_unmetered. - The
code_executiontool answers400with codehosted_tool_unmetered.
On the Gemini route
| What | Path | Answer |
|---|---|---|
| Explicit caching | Path:cachedContents | Answer:402 path_unmetered; implicit caching is automatic and its discount is applied |
| File upload and listing | Path:/upload/…, files | Answer:402 path_unmetered |
| Batch processing | Path:batches, :batchGenerateContent, :batchEmbedContents | Answer:402 path_unmetered |
| Native embeddings | Path::embedContent | Answer:402 path_unmetered; use POST /v1/embeddings |
Imagen (:predict) | Path::predict | Answer:402 path_unmetered |
| Tuned models, corpora, File Search stores | Path:tunedModels, corpora, fileSearchStores | Answer:402 path_unmetered |
| Ephemeral tokens | Path:auth_tokens | Answer:402 path_unmetered |
| Other API versions | Path:Any version but v1beta and v1 (such as v1alpha) | Answer:402 path_unmetered |
| Google Maps grounding | Path:tools: [{googleMaps}] | Answer:402 hosted_tool_unmetered |
| A key in the URL | Path:?key= | Answer:401 |
What is not offered at all
- A rerank endpoint.
- An MCP server for customers.
- OAuth: authentication is by API key only.
- WebRTC for realtime voice; WebSocket only.
- An API to top up the wallet or to create and revoke keys; those are done in the dashboard.
- Models run on infrastructure of our own; every model is run by its own vendor.
A path that is not in this documentation answers 404 with code path_not_found. Every code is in the error reference; any change to this list is recorded in the changelog.
1xAi is an independent gateway. It is not affiliated with, endorsed by, or a reseller for OpenAI, Anthropic, Google, DeepSeek, xAI, Moonshot AI, Z.ai, Alibaba Qwen, MiniMax or Tencent Hunyuan. Requests are relayed upstream rather than answered by models of our own, so each provider's own usage policies apply to what their models will and will not do, and their model availability is theirs to change. Current reachability is published on the status page.