The voice AI landscape: 30 companies, four categories, six pricing models

Voice AI is one of the busiest categories in B2B software right now, and one of the harder ones to get a clear picture of. The companies inside it are usually grouped together as a single market, when in practice they sell quite different things to quite different buyers.

We spend a lot of time researching AI companies, because that is how we build outbound programmes for them. So we took 30 companies from a voice AI market screen and read each one properly: its own website, its own pricing page, its own funding announcements, and the press coverage around it. Everything below was checked on 23 September 2026.

Here is what the landscape looks like.

The short version

• →The 30 companies fall into four categories: speech infrastructure, voice agent platforms, vertical voice applications, and adjacent tools that appear on the same lists.
• →Six pricing models are in active use, from per minute to per seat to per location. Headline rates include different things, so they need a little work before you compare them.
• →The category moves quickly. In the last twelve months one company was acquired by AWS, one went public, and several changed names or broadened what they sell.
• →The companies are spread across four continents, with clusters in India, North America and Europe, and others in Singapore, Japan, South Korea, China, Saudi Arabia and Brazil.

Category 1: speech infrastructure

These companies sell the underlying models and APIs: text to speech, speech to text, and increasingly speech to speech. Their customers are developers and product teams building something on top.

Sarvam builds Indian-language speech models and a sovereign LLM stack from Bengaluru, and raised a $300M Series B in June 2026 with a first close of $234M. Gradium, spun out of the French lab Kyutai, raised a $100M seed and sells real-time speech and live translation from Paris. Cartesia sells its Sonic voices and Ink transcription, and has added managed voice agents. It raised a $64M Series A in March 2025, and its chief executive announced a further $100M in October 2025. Speechmatics in Cambridge has been doing speech to text for years, across 55 languages, with an on-device option and rates published from $0.129 per hour. Smallest.ai builds its own text to speech, speech to text and speech-to-speech models, and raised a $13M Series A in July 2026.

Several serve specific markets. Intella focuses on Arabic dialects across MENA from Riyadh, on a $12.5M Series A raised in September 2025. Typecast, made by Neosapience, sells voice cloning to creators and businesses on credit plans, and listed on Korea's KOSDAQ exchange in September 2026. SinoVoice serves the Chinese market from Beijing. Ultravox sells a speech-to-speech model with an agent platform on top. Hume sells emotion-aware voice models, from $0.07 per minute, alongside a growing data and evaluation business for AI labs. Modulate came from voice safety in games and now also sells audio model APIs to enterprises.

Who buys: engineering teams. What wins: model quality, latency, language coverage and price per unit.

Category 2: voice agent platforms

These companies sell the tooling to build, run and monitor voice agents, usually over the phone. They sit between the models and the business using them, handling orchestration, telephony and integrations.

Vapi is one example, with a $50M Series B in May 2026 and Amazon Ring among its named customers. WIZ.AI serves Southeast Asian enterprises from Singapore, with agents tuned for regional languages. NuPlay AI in Bengaluru, formerly Nurix AI, added chat automation through its acquisition of Verloop.io, announced in July 2026. Voiceflow in Toronto focuses on letting teams design agents together. Ringg AI raised a $15M Series A in August 2026 and has extended from phone calls into WhatsApp and browser agents.

Who buys: product and engineering teams who want to ship an agent without assembling the stack themselves. What wins: reliability, integrations, and how fast a team gets to production.

Category 3: vertical voice applications

These companies pick one industry and one job, and sell a finished outcome rather than a toolkit. The buyer is usually an operations leader rather than an engineer.

Slang.ai answers the phone for restaurants and hospitality groups, and raised a $36M Series B in February 2026. Skit.ai works in debt collections. Prosper raised a $30M Series A led by Andreessen Horowitz in June 2026 for healthcare administration calls, and Outbound AI works on the same problem from Seattle. Concurrence, which renamed from Amigo in September 2026, runs clinical agents that reach patients by phone, SMS and chat. SquadStack runs sales calling for Indian enterprises in nine languages.

Who buys: the line of business, often from a budget that already exists for the work. What wins: depth in one workflow, proof in that industry, and integration with the systems it already runs.

Category 4: adjacent tools

These companies appear on voice AI lists and do genuinely interesting work with audio, but a business shopping for a voice agent would not shortlist them.

David AI licenses multi-speaker audio datasets to model builders, and raised a $50M Series B in October 2025, taking it to $80M according to press reports. Notta and Trint sell transcription, to knowledge workers and newsrooms respectively. Botpress sells an agent builder and has moved into helpdesk software. Voicemod sells a voice changer to gamers and streamers. Async, formerly Podcastle, sells creation tools and now a voice API. CrewAI builds a multi-agent framework, with no speech capability advertised on the pages we reviewed. NLX, an enterprise conversational platform, was acquired by AWS in April 2026 and now sits inside Amazon Connect.

Why this group matters: when a report gives a total for the voice AI market, some of it belongs to businesses like these. Worth knowing which companies sit inside any number you quote.

How these companies charge

Six pricing models are in use across the 30, and the choice says a lot about who a company sells to.

Per minute is the default for voice agents. From the vendors' own pricing pages, checked on 23 September 2026:

The third column is the useful part. These rates look close together, but they cover different things: some include the language model and the telephony, some bill those on top. It is worth rebuilding each one as a cost per conversation before comparing. This table reaches beyond our 30, since several of the best known per-minute vendors were not in the sample.

The other five models in use:

• →Per seat, where a person is the unit. Notta charges $8.17 to $16.67 per seat per month billed annually, and Trint $69 to $79 per seat per month.
• →Per location, which suits multi-site businesses. Slang.ai starts at $399 per location.
• →Per conversation, used by Botpress at $150 a month for 250 conversations, billed annually.
• →Credits, used by Typecast and Gradium, which suits variable creative work.
• →Per hour of audio, used by Modulate and most transcription products.

We found no published rate for 14 of the 30. Most direct you to a sales conversation, which is normal where deals are large and usage varies. Two are different cases: Voicemod shows prices only inside its app, and NLX no longer prices separately now that it sits inside Amazon Connect.

Where the money has gone

Funding has grown quickly, and the totals depend heavily on which companies get counted.

CB Insights, looking at 270 companies focused on voice generation, recorded equity funding of $394M in 2021, $315M in 2022, $264M in 2023 and $2.1B in 2024. For 2026, Newcomer reported more than $7 billion into voice AI startups in the first quarter alone. That piece does not say where the figure comes from or which companies it counts, and it includes at least one company better known for video models.

The two cover different periods and may well count different things, so they cannot be read as one trend line. Treat each as a separate estimate with its own boundary.

Two things help put that in context. In the same CB Insights research, 85 percent of the companies tracked sat in the bottom three of five commercial maturity levels as of March 2025, so much of the category was still early. And for scale, Crunchbase put total AI venture funding at $242 billion in the first quarter of 2026. Voice AI is a small share of AI funding overall, even on the most generous count.

What changed in the last twelve months

Worth tracking if you sell into this category or compete in it.

• →NLX was acquired by AWS, announced 23 April 2026, and now sits inside Amazon Connect.
• →Neosapience, which makes Typecast, went public on KOSDAQ on 21 September 2026, closing its first day at 34,300 won against a 10,000 won offer price.
• →Nurix AI acquired Verloop.io, announced on 9 July 2026, and renamed itself NuPlay AI in August 2026. No closing date has been published.
• →Hume saw its co-founder and chief executive and several engineers join Google DeepMind in a licence-and-hire reported in January 2026, and has built out a data and evaluation business alongside its voice models.
• →Podcastle became Async in January 2026, broadening from podcasting into video, audio and images.
• →Modulate opened the Velma model it built for games as an enterprise API in June 2026.
• →Amigo became Concurrence in September 2026.

What this means if you sell in this category

Four things we would act on, whether you are building one of these companies or selling to them.

1. 1.Say which of the four categories you are in, in your first line. Infrastructure, platform, vertical application and adjacent tool have different buyers and different evaluation criteria. Leaving the buyer to work it out costs you the meeting.
2. 2.Do the pricing arithmetic for your buyer. Since headline rates include different things, the vendor who presents a clear cost per conversation is the easier one to choose.
3. 3.Keep your competitive research current. Several companies here changed name, owner or focus inside a year, so messaging written last quarter may name the wrong competitor.
4. 4.Publish your own facts where you control them. Funding totals, headcounts and customer numbers get repeated from databases into procurement reviews, so it helps to have the correct version on your own site.

The full list

Every company links to its own site. Funding shows the most recent round we could verify from the company's own announcement or from press coverage. Prices are as published on 23 September 2026.

How we did this

Thirty companies from a single market screen, each checked on 23 September 2026 against the company's own website, pricing page and funding announcements first, then press coverage. Where a company publishes no funding total and sources disagree, we left the figure out rather than picking the most quoted one. We could verify a company-stated or press-stated total for 13 of the 30, which is normal for private companies.

Two things we could not establish: any funding announcement for SinoVoice, and Google Cloud's published speech-to-text rate, because that pricing page would not render for us.

If something here is out of date or wrong about your company, tell us and we will correct it with the source.

This is the first in a series of landscape studies. We build them because we run outbound programmes for companies in this category, and the research has to be right before anything is sent. If you would like the underlying data for the segment you sell into, book a GTM call or email hello@spaceout.agency.

Corrections

26 September 2026. We changed the wording on the Verloop.io acquisition twice in one day, so here is the full account. Our research for a later study suggested the 9 July announcement covered an agreement to acquire rather than a completed deal, and we amended the piece to say so. Re-reading the source in full showed it is internally inconsistent: the same article's headline and summary say Nurix has acquired Verloop.io, while a line in its body says the company announced plans to acquire. NuPlay's own site describes the acquisition as complete. No closing date has been published anywhere. The original wording was therefore right, and it has been restored with that last point added. We are leaving this note up rather than deleting it, because a correction log that only records other people's errors is not worth much.

26 September 2026. Async's funding figure was updated from $13.5M to $23.5M. The first figure was the Series A; the second is the total the company states in its own rebrand release.