Speechmatics Pricing 2026: The Training-Discount Catch

TL;DR: Speechmatics pricing starts free with $100 in credit, then Pro runs from $0.129 per hour of speech-to-text with a 20% lock-in discount. The catch: the headline from rate is the cheapest tier, higher-accuracy Enhanced and real-time cost more, and the advertised 33% model-training discount means letting Speechmatics train on your audio.

Last updated: August 2026 · By Sandeep Singh, DailyFix

Speechmatics pricing in 2026: a low from rate, with accuracy tiers and a data-for-discount trade underneath.

Speechmatics is a veteran speech-to-text provider known for accuracy across 56-plus languages, and in 2026 it pushes hard into voice agents with low-latency text-to-speech and a Voice Agent API. Its pricing page, last updated 31 July 2026, looks clean and cheap. But the "from" rate and one opt-in discount hide what you will actually pay and trade. Here is the real math, checked in August 2026.

How does Speechmatics pricing work in 2026?

Speechmatics pricing is credit-based and usage-metered, with three tiers: a Free plan with $100 in credit, a Pro plan from $0.129 per hour with no commitment, and a custom Enterprise plan, per Speechmatics' pricing page updated 31 July 2026. Credits map directly to dollars, where 1 credit equals $1.

You start on $100 of free credit, then add a card to continue at pay-as-you-go rates, billed to the second and invoiced on the 1st of each month for the prior month. The clean part is the credit model: your balance equals your invoice. The part that needs reading twice is that a single "from $0.129/hr" number sits on top of several variables, model accuracy, real-time versus batch, and whether you opt into a data-sharing discount, that each move the real price.

How much does Speechmatics cost per hour?

Speechmatics costs from $0.129 per hour of speech-to-text on Pro, with a 20% discount available for locking in, per Speechmatics' pricing page checked in August 2026. That figure is the entry rate; the higher-accuracy Enhanced model and real-time transcription both cost more than the from price implies.

Speechmatics offers three models: Enhanced for top accuracy, Standard for cost control and fast batch turnaround, and Melia 1 for multilingual audio. Enhanced costs more per hour than Standard, so two customers both "on Pro" can pay very different rates depending on which model they run and whether they stream in real time. The honest way to budget is to price your specific model and mode, for example Enhanced real-time for a live agent, rather than anchoring on the $0.129 headline that represents the cheapest combination.

Plan	Rate	Free credit	Real-time sessions	Commitment
Free	Uses $100 credit	$100	2	None
Pro	from $0.129/hr	$100	50	None (20% off to lock in)
Enterprise	Custom / volume	—	Unlimited	Custom

Prices and limits change often — verify on the official pricing page before deciding.

Is the Speechmatics free credit enough to test?

The Speechmatics free credit of $100, with no card required, is generous enough to run real accuracy tests across languages, try both batch and real-time, and prototype a voice agent before spending your own money. At entry rates, $100 buys hundreds of hours of transcription.

Because the free credit is a one-time $100 grant rather than a monthly reset, treat it as a runway for evaluation, not an ongoing free tier. It is more than enough to benchmark Speechmatics against Deepgram, Gladia, or AssemblyAI on your own audio, which is exactly what you should do before committing. Once the credit is gone, you add a card and move to pay-as-you-go at the standard rate, so use the free window to nail down which model and mode you actually need.

What is the Speechmatics model-training discount?

The Speechmatics model-training discount takes 33% off speech-to-text rates in exchange for opting in to let your audio and transcripts help improve its models, per Speechmatics' pricing FAQ checked in August 2026. It is off by default, reversible anytime, and does not apply to text-to-speech.

This is the catch worth pausing on. A 33% cut is significant, but the price is your data: opting in grants Speechmatics permission to store and use your audio and transcripts for training, though it does not guarantee your data is used. For non-sensitive audio, that trade can be an easy win. For anything covered by confidentiality, healthcare, legal, customer calls, the discount is not really free, and most regulated use cases should leave it off and pay the standard rate. Decide based on how sensitive your audio is, not just the percentage.

Is Speechmatics good for voice agents?

Speechmatics for voice agents is a strong choice, with low-latency speech-to-text and text-to-speech, a dedicated Voice Agent API, 56-plus languages, and 50 concurrent real-time sessions on Pro. As with any transcription provider, it handles the hearing and speaking layers, not the full agent.

A complete voice agent still needs a language model to reason and telephony to place calls, so Speechmatics' per-hour rate is one line item among several. Budget the real-time Enhanced rate if accuracy on live calls matters, because that is the combination a serious agent uses, not the $0.129 floor. For builders who need broad language coverage and enterprise-grade compliance, SOC 2, ISO 27001, GDPR, HIPAA, Speechmatics is one of the more credible voice layers in 2026, provided you price the model and mode you will really run.

Frequently asked questions

Can I use Speechmatics for free?
Yes. Speechmatics gives new accounts $100 in credit with no card required, where 1 credit equals $1. It covers speech-to-text in 56-plus languages and low-latency text-to-speech, with 2 concurrent real-time sessions. When the credit runs out, you add a card to switch to pay-as-you-go at the standard rate.

How much does Speechmatics cost per hour?
Speechmatics Pro starts from $0.129 per hour of speech-to-text, with a 20% discount available for locking in, per Speechmatics' pricing page updated 31 July 2026. That from rate is the lowest tier; the higher-accuracy Enhanced model and real-time transcription cost more than the entry figure suggests.

What is the Speechmatics model-training discount?
The model-training discount takes 33% off Speechmatics speech-to-text rates in exchange for opting in to let your audio and transcripts help improve its models. It is off by default, reversible at any time, and applies only to future usage. It does not apply to text-to-speech, and opting in does not guarantee your data is used.

Does Speechmatics offer volume discounts?
Yes. Speechmatics applies volume discounts automatically on billable usage above 500 hours per speech-to-text type in a month, with 20% off the hours beyond 500. Additional discounts start from 24,000 hours of usage a year, arranged with sales. Below 500 hours a month you pay the standard rate.

What is the difference between Speechmatics Enhanced and Standard?
Enhanced is Speechmatics' highest-accuracy model across all languages, while Standard trades some accuracy for cost control and faster batch turnaround. There is also Melia 1 for multilingual audio in a single transcript. Enhanced costs more per hour than Standard, so the model you choose changes your bill as much as the plan does.

Is Speechmatics good for voice agents?
Yes. Speechmatics offers low-latency speech-to-text and text-to-speech built for voice agents, plus a dedicated Voice Agent API, with 50 concurrent real-time sessions on Pro. Budget the real-time and Enhanced rates rather than the headline from price, and remember it is the transcription and voice layer, not a full agent stack.

Bottom Line

Speechmatics is one of the most accurate, broadly multilingual speech APIs in 2026, and its $100 free credit makes it easy to benchmark risk-free. Use the free credit to price the exact model and mode you need, Enhanced or Standard, real-time or batch, rather than trusting the $0.129 from rate. Take the 33% training discount only for non-sensitive audio; leave it off for anything confidential. Choose Speechmatics if you need language breadth and enterprise compliance, and budget it as one layer of your voice-agent stack.

Sources

Speechmatics — Official pricing (updated 31 July 2026)
Speechmatics — Documentation

About the author — Sandeep Singh. Sandeep builds and tests AI tools and automations daily, and runs DailyFix to show solopreneurs which ones actually save time.

Sandeep Singh Founder & Editor, DailyFix

Sandeep Singh is the founder and editor of DailyFix. He served in the Border Security Force for five years before moving into software, where he now builds and tests AI automations for solopreneurs and small teams — mostly in n8n, Make, and Zapier, plus voice agents on Vapi and Synthflow. He runs DailyFix on the same principle he uses in his own work: no tool goes into a post unless he'd actually put it into a client build. Every review, prompt and workflow on the site is tested end-to-end on real projects before it's written up. Alongside DailyFix, Sandeep ships SwiftSite Studio. Recent focus: AI receptionists for contractors, and turning ranking-audit output into repeatable, one-person publishing systems. Reach him on GitHub.

#Speechmatics pricing #Speechmatics cost per hour #speech-to-text API #Speechmatics for voice agents #STT pricing 2026 #Speechmatics model training