AssemblyAI sync API 120s 40MB caps 30s deadline

Not specified by the original author.

TL;DR

Use AssemblyAI's sync API only for short clips under 120 seconds and 40 MB: pass a file, not a URL, set the model with the X-AAI-Model header, and keep each request under the 30-second deadline. For longer audio, public URLs, or features like speaker labels and chapters, use the async flow.

Steps

1. Use AssemblyAI's sync API only for short clips under 120 seconds and 40 MB: pass a file, not a URL, set the model with the X-AAI-Model header, and keep each request under the 30-second deadline. For longer audio, public URLs, or features like speaker labels and chapters, use the async flow. Pre-warm with the warm endpoint to skip the connection handshake on the critical path.

When to use

You are seeing this: Pre-warm with the warm endpoint to skip the connection handshake on the critical path. Use this skill when you run into "AssemblyAI sync API: 120s/40MB caps, 30s deadline, files not URLs".

When not to use

If your error message or symptom does not match what is described above, this is probably not your fix. Search for your exact error text instead of forcing this one to fit.

Versions

No specific versions are mentioned in the source material, so treat the fix as generally applicable and check the examples against whatever you have installed.

Why this happens

The original report does not dig into a root cause. It documents the symptom and the fix that resolved it.

Source: Official AssemblyAI docs (API reference, assemblyai/assemblyai-skill repo): documents the sync API constraints agents hit. The sync endpoint accepts only files, not URLs: a local path, raw bytes, or a file object. Audio is limited to 80 ms to 120 s and 40 MB; for longer audio use the async REST API. Each request has a 30-second deadline that returns 504 inference_timeout. The model is selected via the X-AAI-Model header, never in the request body, and unknown config fields are rejected with a 400.

Verified resolution: Use AssemblyAI's sync API only for short clips under 120 seconds and 40 MB: pass a file, not a URL, set the model with the X-AAI-Model header, and keep each request under the 30-second deadline. For longer audio, public URLs, or features like speaker labels and chapters, use the async flow. Pre-warm with the warm endpoint to skip the connection handshake on the critical path.

Source URL: https://github.com/assemblyai/assemblyai-skill/blob/HEAD/skills/assemblyai/references/api-reference.md

Not specified by the original author.

TL;DR

Use recognize for clips under 60 seconds and longrunningrecognize beyond that. Poll the returned operation name until done is true, or pass output_config with a GCS URI so the transcript lands in your bucket without polling.

Steps

1. Use recognize for clips under 60 seconds and longrunningrecognize beyond that. Put audio longer than a minute in a GCS bucket first; a local file over 60 seconds errors even on async calls. Poll the returned operation name until done is true, or pass output_config with a GCS URI so the transcript lands in your bucket without polling.

When to use

You are seeing this: Put audio longer than a minute in a GCS bucket first; a local file over 60 seconds errors even on async calls. Use this skill when you run into "Speech-to-Text long audio: sync caps at 60s, async needs GCS for longer files".

When not to use

If your error message or symptom does not match what is described above, this is probably not your fix. Search for your exact error text instead of forcing this one to fit.

Versions

No specific versions are mentioned in the source material, so treat the fix as generally applicable and check the examples against whatever you have installed.

Why this happens

The original report does not dig into a root cause. It documents the symptom and the fix that resolved it.

Source: Official docs (Cloud Speech-to-Text, long audio): documents the recognition-mode gotchas agents hit. Synchronous recognize caps at 60 seconds of audio. For anything longer use longrunningrecognize, which handles up to 480 minutes, but local audio files are still capped at 60 seconds even for async jobs; longer audio must live in a Cloud Storage bucket. Poll the operations endpoint until done is true. Results stay retrievable for 5 days, or set output_config to have the transcript written to GCS directly.

Verified resolution: Use recognize for clips under 60 seconds and longrunningrecognize beyond that. Put audio longer than a minute in a GCS bucket first; a local file over 60 seconds errors even on async calls. Poll the returned operation name until done is true, or pass output_config with a GCS URI so the transcript lands in your bucket without polling.

Source URL: https://cloud.google.com/speech-to-text/docs/async-recognize

Not specified by the original author.

TL;DR: The synchronous process endpoint timed out - 400 pages of OCR never fit in its deadline. Switch to batch processing: drop the file in Cloud Storage, fire a batchProcess request, and poll the long-running operation. That path is built for exactly this size.

DEADLINE_EXCEEDED

1. Confirm you are calling the synchronous process method on a large document. Expected: yes - that combination is the classic trigger.

2. Upload the MSA to a Cloud Storage bucket the processor's service account can read. Expected: listing the bucket shows the file.

3. Submit a batchProcess request pointing at the Cloud Storage input and an output prefix. Expected: the API returns a long-running operation name immediately.

4. Poll the operation until done, then read the sharded JSON output from the output prefix. Expected: one JSON shard per chunk, with full OCR for all 400 pages.

5. If you must stay synchronous (small follow-up docs), split the MSA into page ranges under the sync page cap and merge results. Expected: each chunk processes within the deadline.

Use this when

• processDocument returns DEADLINE_EXCEEDED on a long contract
• the same processor handles short docs fine
• you are tempted to "just raise the timeout" - the server-side limit still applies

Not for this skill when

• the error hits a 5-page doc (check processor health, quotas, or your network instead)
• the error is INVALID_ARGUMENT (bad MIME type or corrupt file - different fix)
• batch jobs also time out (check quotas and processor version)

Variant phrasings

document ai DEADLINE_EXCEEDED large pdf

google document ai timeout 400 pages

document ai batch process vs synchronous process

Why it happens

Synchronous processing holds one call open for the whole job and enforces a fixed deadline. A 400-page MSA needs minutes of OCR; the deadline fires first. Batch processing runs asynchronously with no such deadline and writes results to storage as it goes.

Edge cases

• Batch output is sharded - your reader must concatenate shards in order.
• Raising the client-side deadline only helps marginal cases; the service still caps sync work.
• Check the processor's max page count; some processors cap below 400 even for batch - split if so.
• Batch costs and latency differ from sync; budget both before switching a whole pipeline.

Provenance

Resolved from the public thread: https://vectle.com/posts/pstvrG2Hue3DwYP3afneQkZw

Resolved from the public thread: https://vectle.com/posts/pstvrG2Hue3DwYP3afneQkZw

Thread query: Google Document AI "DEADLINE_EXCEEDED" on a 400-page MSA

TL;DR: