10 November 2026 · Site Methodology & Corrections · 5 min read
I was updating our transcription calculator's feature-add-on table today and pulled AssemblyAI's live pricing page to double-check a diarization fee. Instead I noticed the base rate: Universal-2 async transcription is $0.15 an hour. That's not what any of our own content says AssemblyAI costs. Our July post names Deepgram the cheapest option at $0.26/hr and puts AssemblyAI last, at $0.37/hr. Five months later, that ranking is backwards.
Pulled live from deepgram.com/pricing and assemblyai.com/pricing on 2026-11-10, batch/async tiers only (not streaming):
| Provider & tier | Per minute | Per hour |
|---|---|---|
| AssemblyAI Universal-2 (async) | $0.0025 | $0.15 |
| AssemblyAI Universal-3.5 Pro (async) | $0.0035 | $0.21 |
| Deepgram Nova-3 (batch) | $0.0043 | $0.258 |
| OpenAI Whisper (API) | $0.006 | $0.36 |
AssemblyAI's cheaper tier, Universal-2, now undercuts Deepgram by 42% per hour of audio. Even its newer, more accurate Universal-3.5 Pro tier beats Deepgram by about 19%. Both numbers come straight off AssemblyAI's own pricing page, not a cached snapshot — I fetched it fresh while writing this.
The post we published on 2026-07-26 priced AssemblyAI's "Universal" model at a flat $0.37/hr and called Deepgram Nova the cheapest of the three options compared, at $0.26/hr. That was a correct read of AssemblyAI's pricing at the time — Universal-2 and Universal-3.5 Pro as separate, cheaper async tiers didn't exist yet under those names and rates. The problem is the post never got revisited after AssemblyAI restructured its tiers. It's still live, still ranking Deepgram first, and still the page Google shows for "speech to text cost per hour."
Our actual transcription cost calculator is not wrong here — it already lists AssemblyAI batch at $0.0035/min, matching Universal-3.5 Pro. What it's missing is the even cheaper Universal-2 line, so even our calculator understates how far ahead AssemblyAI now is for the budget-conscious case.
Take a podcast network transcribing 2,000 hours of audio a month — a realistic mid-size workload:
| Provider & tier | 2,000 hrs/mo cost | vs. Universal-2 |
|---|---|---|
| AssemblyAI Universal-2 | $300 | — |
| AssemblyAI Universal-3.5 Pro | $420 | +$120/mo |
| Deepgram Nova-3 | $516 | +$216/mo |
| OpenAI Whisper | $720 | +$420/mo |
$216/month between the cheapest and what our own July post called the cheapest. Over a year that's roughly $2,600 — the kind of gap that's worth a one-line config change, not a migration project, since both providers expose a similar REST-upload-and-poll API shape.
Price isn't the whole story. Universal-2 is AssemblyAI's older, more established model; Universal-3.5 Pro is the newer one with better accuracy on accents, overlapping speech and domain-specific vocabulary, which is presumably why it costs 40% more. If your audio is clean single-speaker English, Universal-2's $0.15/hr is very likely good enough. If you're transcribing noisy multi-speaker calls where word-error-rate actually affects downstream decisions, the extra $0.06/hr for Universal-3.5 Pro is cheap insurance — it's still under Deepgram either way.
Added a correction note to the July post pointing here, and adding a Universal-2 row to the calculator's provider list so the cheapest real option actually shows up as an option, not just as a number buried in a blog post. Same lesson as every other correction post in this series: a provider changing its pricing page doesn't announce itself, and the only way to catch it is to keep re-checking the numbers you already published instead of assuming they age well.
New to transcription API pricing? Start with the free API-cost guides.
Deploy it yourself: DigitalOcean — $200 free credit ↗ · Hostinger VPS ↗
Prices pulled live from deepgram.com/pricing and assemblyai.com/pricing on 2026-11-10, async/batch tiers only. Reference figures, not live quotes -- confirm current terms on each provider's own pricing page before budgeting at volume.