> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.beatapi.io/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.beatapi.io/_mcp/server.

# A long text request disconnected. Must I generate it again?

> Save client errors and partial text before rerunning a dropped text call; route long requests to https://direct.beatapi.io/v1 with streaming enabled instead of polling /v1/tasks.

A disconnected text call does not always require an immediate blind retry. First capture the client error, HTTP status, and any partial response text to distinguish a network timeout, an interrupted stream, and a gateway or model error.

## Prevent timeouts on long text replies

* **Use the direct text endpoint**: Send text requests to `https://direct.beatapi.io/v1` rather than `https://api.beatapi.io/v1`. The direct endpoint avoids Cloudflare's 100-second wait limit on long non-streaming responses while keeping the same API key, models, and prices.
* **Enable streaming**: Set `"stream": true` on Chat Completions, Responses, or Messages so tokens arrive incrementally via server-sent events instead of waiting for one full payload.

## Handle partial output and retries safely

* **Quarantine incomplete text**: When only half of a summary or document arrives, flag the record as incomplete and block incomplete text from downstream publishing or final drafts.
* **Send a new text request when needed**: Text endpoints never create asynchronous jobs, so `GET /v1/tasks/{task_id}` applies only to media, workflow, effect, or data task IDs, not text completions.
* **Retry selectively**: Retry only `429`, `500`, and `503` failures with backoff; do not resend `400`, `401`, `402`, `403`, `404`, or `413` requests unchanged.

## Contracts and related questions

Before sending a paid request, inspect the model contracts for free: `model:gpt-6-luna`, `model:gpt-image-2.5-flare-std`.

[Back to AI Models questions](/faq/models).