AssemblyAI Pricing 2026: Universal Models per Hour and Add-On Costs

Short answer: AssemblyAI charges per hour of audio with no subscription. Pre-recorded transcription costs $0.15 an hour on Universal-2 and $0.21 an hour on Universal-3.5 Pro, its most accurate batch model; realtime runs from $0.15 to $0.45 an hour. The free tier covers up to 185 hours of pre-recorded or 333 hours of streaming audio, no card needed. Verified on AssemblyAI's pricing page, October 2026.

Those base rates are low, but almost nobody pays only the base rate. Speaker labels, key terms, redaction and summaries are each billed per hour on top, and on the realtime side speaker diarization nearly doubles the cheapest model's price. Below is what 1,000 hours actually costs for four common setups.

AssemblyAI speech-to-text prices at a glance

Rates read on AssemblyAI's pricing page on October 30, 2026. Custom pricing with higher rate limits is available through sales.

Dupple data: Toolradar, our software directory, counts 176 transcription tools as of October 30, 2026; 117 of them have a free plan, and the 67 that publish a paid price have a median cheapest plan of $12 a month (Toolradar's transcription ranking). At AssemblyAI's Universal-2 rate, $12 buys 80 hours of audio (our arithmetic), which is why per-hour APIs beat subscriptions for anyone building transcription into a product.

What add-ons cost on top of transcription

Add-on prices differ between the batch and realtime APIs, so check the column you will actually use:

Add-on rates from AssemblyAI's pricing page, October 30, 2026. Batch PII text redaction is listed under Guardrails.

Realtime speaker diarization costs six times the batch rate. On Universal-Streaming it takes the price from $0.15 to $0.27 an hour, an 80% jump (our arithmetic). If you only need speaker labels after the call, transcribe the recording in batch instead.

Medical mode is the other expensive switch. It costs as much per hour as Universal-2 itself, so a medical scribe product pays roughly double the headline rate before any other feature.

Speech understanding and guardrails prices

These models run on a finished transcript and are billed per hour of audio:

Rates from AssemblyAI's pricing page, October 30, 2026.

Low effort is the default for speaker identification and summarization. Medium effort, which AssemblyAI suggests for long, multilingual or detail-critical audio, costs 3.5 to 5 times more, so switch it on per job rather than account-wide.

For anything these models do not cover, AssemblyAI's LLM Gateway runs language models on your transcripts at per-token prices, from $0.05 per million input tokens on its cheapest listed models. Those prices assume global routing; AssemblyAI says US or EU in-region routing costs 10% more.

What 1,000 hours of audio costs on AssemblyAI

Here are four common setups, each priced at 1,000 hours. All totals are our arithmetic on AssemblyAI's printed rates.

Add-ons make up 12% to 40% of the hourly price across these four setups (our arithmetic). Budget for the stack you will run, and cut features you do not read: a summary nobody opens still costs money on every hour.

The budget row is the cheapest serious setup, and it is cheaper than it looks for Universal-2 users because keyterms prompting is included on that model. Upgrading the same 1,000 hours to Universal-3.5 Pro with diarization costs $60 more, a sensible price for the accuracy jump on difficult audio.

Is AssemblyAI free?

It has a generous free tier rather than a free plan. A new account can transcribe up to 185 hours of pre-recorded audio and up to 333 hours of streaming audio, with no credit card. The free tier is limited to 5 new streaming connections per minute, which is enough for development but not for a live product.

After that, the account moves to pay-as-you-go. For a small app or a research project, the free hours can cover months of real use. If you want a finished app rather than an API, see our list of free transcription software.

Hidden costs to budget for

These come from the FAQ and plan notes on AssemblyAI's pricing page, read October 30, 2026.

• Multichannel audio bills per channel. A one-hour stereo file is billed as two hours of transcription. If your recordings put each speaker on a separate channel, either accept the double rate or mix down and use diarization instead.
• Realtime features cost more than batch. Diarization and PII text redaction are both priced higher on the realtime API, so a live product pays more for the same transcript features.
• LLM Gateway in-region routing adds 10%. Data residency requirements raise every token price.
• Monthly billing in arrears. Invoices arrive at the start of each month for the previous month's usage, with no minimum or contract on pay-as-you-go. Track usage during the month so a runaway job does not surprise you.

Concurrency, by contrast, costs nothing extra. Pay-as-you-go accounts start at 100 new streaming sessions per minute, and the limit rises by 10% whenever you use 70% or more of it, with no ceiling and no fee.

How AssemblyAI pricing changed in 2026

An archived copy of AssemblyAI's pricing page from January 10, 2026 shows a different lineup:

• Top accuracy got cheaper and multilingual. In January, the most accurate batch model was Slam-1 at $0.27 an hour, English only and in beta. Today Universal-3.5 Pro holds that spot at $0.21 an hour across 18 languages, 22% less (our arithmetic).
• The base rate held. Universal (now Universal-2) and Universal-Streaming cost the same in January as they do now.
• The free hours did not change. January's page offered the same 185 pre-recorded and 333 streaming hours.
• New premium tiers arrived. Universal-3.6 Pro Realtime, the Sync API, the Dictation API and a text-to-speech model at $15 per million characters are all additions since January.

Trackers that still quote Slam-1 or an older Universal-3 Pro name are out of date.

AssemblyAI text-to-speech pricing

AssemblyAI-TTS-1, its first voice model, costs $15 per million input characters, with cloned voices and EU processing at the same price. It streams in English, Spanish, French, German, Italian and Portuguese, includes a library of 4,000+ voices and instant voice cloning, and needs no spend commitment for concurrency. That makes AssemblyAI a possible single vendor for both ends of a voice agent; for the full field, see our ranking of the best AI voice agents.

Which AssemblyAI model should you pick?

Pick Universal-2 for high-volume English batch work where cost dominates, Universal-3.5 Pro for meetings, interviews and calls where names and jargon must be right, and Universal-Streaming for live captions on a budget. Universal-3.6 Pro Realtime is for voice agents where every misheard word breaks the conversation.

For how AssemblyAI ranks against other transcription APIs and apps, see the best AI transcription tools; for analytics products built on top of transcripts, see AI speech analytics tools. Toolradar keeps an AssemblyAI pricing profile with plan details.

FAQ

How we checked these prices

We read AssemblyAI's pricing page, including its FAQ, on October 30, 2026, and compared it with an archived copy of the same page from January 10, 2026. Scenarios use 1,000 hours of audio; every total is our arithmetic on printed rates. We will re-check these rates within 60 days.

Cite this: Dupple, "AssemblyAI Pricing 2026: Universal Models per Hour and Add-On Costs", November 2026.

Transcription is cheap until the features stack up. Want more cost breakdowns like this for AI at work? Try Dupple X →

Related

• Best AI transcription tools: AssemblyAI against other APIs and apps.
• Best AI speech analytics tools: analytics built on transcripts.
• Best free transcription software: apps with real free plans.
• AssemblyAI on Toolradar: the listing with plan details.
• Dupple learning hub: more guides on AI tools and costs.