Published 2026-09-22 · Tested 2026-09-22

AssemblyAI

A-

AssemblyAI received 10 PASS votes and passed 5 of five agent surface checks. The clearest finding came from the recover from a 429 task.

Panel: GPT 5.6 Sol, Opus 5, DeepSeek v4F Battery: v1 Read as markdown (opens in a new tab)

Three AI models, GPT 5.6 Sol, Claude Opus 5, and DeepSeek v4 Flash, each read AssemblyAI’s public documentation independently and attempted five first-hour developer jobs: transcribe the first audio file, find the exact limits, recover from a 429, verify a webhook, use the Python SDK.

No accounts, API calls, or code execution were used. Every verdict came from public pages and every published quotation passed a live verification check.

Freshness

How rechecks work
Category
Voice & speech
Tested
Quotes verified
Surface rechecked
Not yet rechecked
AssemblyAI Mintlify · published
A-

92.5% · 74/80 · AI Agent Readiness Score · reading 30 pts · surface 50 pts

llms.txt PASS
llms-full.txt PASS
markdown mirror PASS
MCP server PASS
docs AI PASS
Task GPT 5.6 SolOpus 5DeepSeek v4F Consensus
Transcribe the first audio file PASSPASSPASS PASS
Find the exact limits PARTIALPASSPASS PASS
Recover from a 429 FAILPARTIALPARTIAL PARTIAL
Verify a webhook PASSPASSPASS PASS
Use the Python SDK PASSPARTIALPASS PASS

docs platform: Mintlify (unscored) · verified 2026-09-22

What to fix first

This fix could add up to 4 points to the AI Agent Readiness Score. The list ranks each fix by the points it would add. How the ranking works

  1. 1
    +4 points Recover from a 429 PARTIAL

    Found: The common errors table lists 429 with no fix, the rate-limits page says the HTTP ceiling returns 403, and the backoff remedy sits only under LLM Gateway.

    Fix: Add the exponential backoff with jitter remedy to the transcription common errors guide, and state which condition returns 429 rather than 403.

    Evidence: www.assemblyai.com/docs/llm-gateway/troubleshooting (opens in a new tab)

What the docs get right

  • Transcribe the first audio file: 3 PASS votes. The quickstart links to the dashboard, installs the Python SDK, runs a sample file, and prints the transcript text.
  • Verify a webhook: 3 PASS votes. The guide documents URL setup, POST delivery, 2xx acknowledgment, retries, payload retrieval, and a custom authentication header.
  • 5 of 5 agent surface checks. Present: llms.txt, llms-full.txt, markdown mirrors, an MCP server, docs AI.

Transcribe the first audio file

PASS

PASS consensus from 3 PASS.

The quickstart links to the dashboard, installs the Python SDK, runs a sample file, and prints the transcript text. One quickstart gives the key location, the pip install, a five-line Python script against a hosted sample file, and the equivalent raw HTTP flow in the same page. One clear path from API key (dashboard) to working transcription via SDK using a hosted sample file.

Find the exact limits

PASS

PASS consensus from 2 PASS, 1 PARTIAL.

The rate-limit page says excess HTTP traffic returns 403, but the common-errors page associates excess requests with 429. One page carries both numbers (5 parallel transcriptions free, 200+ paid, plus the 20,000-per-five-minutes HTTP ceiling) and every other page that restates them uses identical figures. Rate limits documented consistently: 5 parallel transcriptions (free), 200+ (paid), plus an HTTP ceiling of 20,000 requests per 5 minutes returning 403.

Recover from a 429

PARTIAL

PARTIAL consensus from 2 PARTIAL, 1 FAIL.

The 429 entry gives no correction, while the actionable HTTP-limit guidance describes a 403 response. A complete 429 remedy exists only for LLM Gateway; on the transcription side the error table lists 429 with no fix while the rate-limits page states the HTTP ceiling returns 403, so a 429 on /v2/transcript has no documented cause or correction. The generic error table identifies 429 as 'Too many requests,' but the rate-limits page states the HTTP ceiling breach returns 403, not 429. The fix (honour Retry-After, exponential backoff) appears only in the agent instructions page, not in the main error handling guide.

Verify a webhook

PASS

PASS consensus from 3 PASS.

The guide documents URL setup, POST delivery, 2xx acknowledgment, retries, payload retrieval, and a custom authentication header. The page covers setting webhook_url, the delivery payload shape, the 10-second/10-attempt retry policy, the two static sender IP addresses, and the webhook_auth_header_name and webhook_auth_header_value shared-secret pair. Complete end-to-end guide: set webhook_url and auth header on the transcription request, validate the custom header on receipt, return 2xx within 10 seconds, with documented retry behaviour and static source IPs.

Use the Python SDK

PASS

PASS consensus from 2 PASS, 1 PARTIAL.

The Python example installs assemblyai, transcribes the same sample URL, and the REST example submits that URL as audio_url. The SDK example matches the REST form shown on the same quickstart page, but the docs teach two incompatible Python import styles side by side, so an agent copying from the wrong page writes code against a different API surface. Official Python SDK on PyPI with a minimal working example; the pattern (set api_key, create Transcriber, call transcribe(audio_url)) matches the REST endpoint (POST /v2/transcript with audio_url) consistently across SDK and HTTP examples.

The receipt

Implement exponential backoff with jitter when you receive a 429.

A complete 429 remedy exists only for LLM Gateway; on the transcription side the error table lists 429 with no fix while the rate-limits page states the HTTP ceiling returns 403, so a 429 on /v2/transcript has no documented cause or correction.

Agent surface notes

Initialize returned JSON-RPC protocol 2025-06-18 and serverInfo name AssemblyAI.

The hydrated docs DOM exposes an Ask Assistant button with aria-labels "Toggle assistant panel" and "Ask a question...".

Show the score

AI Agent Readiness Score 92.5%, grade A-

Paste this into a readme:

[![AI Agent Readiness Score 92.5%](https://docsforagents.com/badge/assemblyai.svg)](https://docsforagents.com/reports/assemblyai-docs-ai-agent-readiness/)

Method note

This is a reading test of public documentation, not an execution test. No accounts were created and no API calls were run. The AI Agent Readiness Score counts fifteen reading votes at PASS 2, PARTIAL 1, and FAIL 0, for 30 possible points. Five agent surface checks add 10 points each. The total is 80. Consensus chips show each row majority and do not affect scoring. The panel split on 3 of five tasks. Quotes shown here were re-fetched and confirmed verbatim on 2026-09-22.

Read the full methodology

Put another docs site through the battery.

Nominate a docs site