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

Deepgram

B-

Deepgram received 10 PASS votes and passed 4 of five agent surface checks. The clearest finding came from the find the exact limits 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 Deepgram’s public documentation independently and attempted five first-hour developer jobs: transcribe the first pre-recorded file, find the exact limits, recover from a 429, verify a Callback delivery, 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
Deepgram Fern · published
B-

81.2% · 65/80 · AI Agent Readiness Score · reading 30 pts · surface 50 pts

llms.txt PASS
llms-full.txt FAIL
markdown mirror PASS
MCP server PASS
docs AI PASS
Task GPT 5.6 SolOpus 5DeepSeek v4F Consensus
Transcribe the first pre-recorded file PASSPASSPASS PASS
Find the exact limits PARTIALPARTIALPARTIAL PARTIAL
Recover from a 429 PARTIALPARTIALPASS PARTIAL
Verify a Callback delivery PASSPASSPASS PASS
Use the Python SDK PASSPASSPASS PASS

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

What to fix first

These 3 fixes could add up to 15 points to the AI Agent Readiness Score. The list ranks each fix by the points it would add. How the ranking works

  1. 1
    +10 points llms-full.txt check failed

    Found: HTTP 301 redirecting to llms.txt, so no separate full-text file exists.

    Fix: Publish llms-full.txt at the docs root with the full text of every docs page in one plain-text file.

    Evidence: developers.deepgram.com/llms-full.txt (opens in a new tab)

  2. 2
    +3 points Find the exact limits PARTIAL

    Found: The pre-recorded guide states 100 concurrent requests per project, while the API Rate Limits reference states 50 on every self-serve plan and region.

    Fix: Correct the pre-recorded audio guide to match the per-plan, per-region concurrency table in the API Rate Limits reference, or state which value applies.

    Evidence: developers.deepgram.com/docs/pre-recorded-audio (opens in a new tab)

  3. 3
    +2 points Recover from a 429 PARTIAL

    Found: The errors page defines 429 as a per-time-window limit, while the concurrency guide says Deepgram limits only concurrent requests, not requests per time span.

    Fix: Rewrite the 429 entry on the errors page to describe a concurrency limit, and repoint its broken #rate-limits link to the API Rate Limits reference.

    Evidence: developers.deepgram.com/docs/working-with-concurrency-rate-limits (opens in a new tab)

What the docs get right

  • Transcribe the first pre-recorded file: 3 PASS votes. The linked key guide and pre-recorded guide provide one account-to-key-to-cURL path for remote or local audio.
  • Verify a Callback delivery: 3 PASS votes. The callback guides document the POST receiver, Listen request, asynchronous response, retries, and Basic Auth as the reliable origin check.
  • 4 of 5 agent surface checks. Present: llms.txt, markdown mirrors, an MCP server, docs AI. Missing: llms-full.txt.

Transcribe the first pre-recorded file

PASS

PASS consensus from 3 PASS.

The linked key guide and pre-recorded guide provide one account-to-key-to-cURL path for remote or local audio. One linked chain runs from signup to key to a runnable curl call against https://api.deepgram.com/v1/listen?model=nova-3&smart_format=true, with no guessing required. One clear path: create account via Console, get API key, then follow the pre-recorded audio guide with cURL or SDK examples.

Find the exact limits

PARTIAL

PARTIAL consensus from 3 PARTIAL.

The plan tables give per-region caps, but they state 50 pre-recorded requests where the pre-recorded guide states 100. The rate limits reference gives a complete per-plan, per-region, per-model table, but the pre-recorded guide states 100 concurrent pre-recorded requests where the reference states 50 on every self-serve plan and region. Pre-recorded page says 100 concurrent requests but the API Rate Limits page says 50 for Nova-3 pre-recorded on Pay as You Go. Pages contradict each other.

Recover from a 429

PARTIAL

PARTIAL consensus from 1 PASS, 2 PARTIAL.

The concurrency guide names the cause and recommends queues and backoff, but the errors page also defines 429 by request volume over time. The error reference names the payload and recommends exponential backoff, but its status-code list calls 429 a per-time-window limit while the concurrency guide denies any per-time-window limit, and the reference's own rate-limit link points at a #rate-limits anchor that no longer exists on the redirected pre-recorded page. Cause named, response format shown (TOO_MANY_REQUESTS error code), and exponential-backoff retry recommended with link to dedicated guide.

Verify a Callback delivery

PASS

PASS consensus from 3 PASS.

The callback guides document the POST receiver, Listen request, asynchronous response, retries, and Basic Auth as the reliable origin check. The page documents the callback parameter, the permitted ports and protocols, the POST delivery, the ten-retry policy at 30 second intervals, and both authentication methods, including the caveat that dg-token is not guaranteed on every request. Full callback path documented: query parameter, server setup, retry behavior (10 attempts at 30s intervals), and two authentication methods (Basic Auth and dg-token header).

Use the Python SDK

PASS

PASS consensus from 3 PASS.

The home page links the official SDK, and the guide provides installation plus a complete binary-audio example matching the REST local-file form. pip install deepgram-sdk python-dotenv plus the DeepgramClient example uses the same model=nova-3 and smart_format=true parameters as the curl form on the same page, though /docs/python-sdk redirects to the home page and the only Python SDK link is to GitHub. Official SDK on GitHub with clear README, pip install, and minimal working examples for URL and file transcription that match the API form shown in the developer docs.

The receipt

Rate limits: Up to 100 concurrent requests per project for Nova, Base, and Enhanced models. For full details, see API Rate Limits.

The plan tables give per-region caps, but they state 50 pre-recorded requests where the pre-recorded guide states 100.

Agent surface notes

Initialize returned JSON-RPC protocol 2025-06-18 and serverInfo name fern-docs-mcp-server.

The hydrated docs DOM exposes an Ask AI control with aria-labels "Next: Ask AI" and "Show voice assistant".

Show the score

AI Agent Readiness Score 81.2%, grade B-

Paste this into a readme:

[![AI Agent Readiness Score 81.2%](https://docsforagents.com/badge/deepgram.svg)](https://docsforagents.com/reports/deepgram-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 1 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