How we checked: every status below comes from the vendor's own site, an archived copy of it, a regulator, or the acquirer's announcement. Where a date or detail comes only from the press, we say so. Prices come from each vendor's pricing page, read November 11, 2026. We did not run these APIs ourselves.
Voice AI vendors that shut down or changed, 2025-2026#
| Vendor | What happened | Source | Status today |
|---|---|---|---|
| PlayHT / PlayAI | Meta acquired PlayAI in July 2025 (a Meta spokesperson confirmed it to Bloomberg). By January 6, 2026 the play.ht homepage read "We have shut down the service." | Archived play.ht, Bloomberg | Closed. play.ht no longer resolves. |
| LMNT | Site replaced by "LMNT has shut down. Our speech generation journey has come to an end." No date, successor or refund terms given. | lmnt.com | Closed. |
| Hume Octave TTS and EVI | Sunset notice dated October 2, 2026: access ends November 13, 2026 at 12:01 a.m. EST; account data deleted afterwards. | Hume changelog, archived | Ending Nov 13. Hume continues with an Audio Decisions API. |
| Vocode | Open-source voice-agent library. Last commit to its main repository was November 15, 2024; vocode.dev now redirects to the GitHub organisation. | GitHub | Dormant. Code still available (not archived). |
| Air AI | FTC sued in August 2025 over earnings and refund claims; settlement announced March 24, 2026 bans the company and owners from selling business opportunities. The air.ai domain now hosts an unrelated defense company. | FTC | No Air AI voice product found. |
| Synthflow | Dropped self-serve plans: now "sold as an annual enterprise agreement" with no free plan, trial or per-minute rate. | Synthflow pricing | Active, enterprise only. |
| Ultravox | Acquired by Inworld (announced September 30, 2026). Inworld says existing voice IDs keep working. | Inworld | Active, $0.05 a minute. |
| Gladia | Acquisition by OVH Groupe completed July 31, 2026. Gladia says its API, contracts and pricing are unchanged. | OVHcloud | Active. |
Two patterns stand out. Small voice API companies are being bought for their teams (PlayAI by Meta) or are closing outright (LMNT), and in both cases customers got little or no public notice we could find. Hume, by contrast, published a dated notice six weeks ahead. If a vendor in your stack is small and quiet, keep your voice source audio and your integration behind an abstraction you control.
What PlayHT and LMNT users need from a replacement#
PlayHT was used two ways: as a voiceover studio for creators, and as a low-latency API for apps and voice agents. LMNT was API-only, aimed at real-time apps, with fast voice cloning. So the replacement criteria are latency, voice cloning, price per hour of audio and language coverage. For voiceover studio work rather than an API, our ElevenLabs alternatives guide covers the creator tools.
TTS APIs compared, converted to cost per hour#
Vendors bill per character or per minute of audio. To compare them, we converted to cost per hour of generated speech at 800 characters per minute (48,000 characters an hour), close to Cartesia's own estimate of 750 to 800. Inworld and Rime assume about 1,000 characters a minute, which would raise every per-character figure by a quarter.
| TTS API | Published price | Per hour (our conversion) | Languages (vendor) | Voice cloning | Time to first audio (Coval) |
|---|---|---|---|---|---|
| Inworld TTS-2 Flash | $15 per 1M chars on demand; $7 on Growth | $0.72; $0.34 | 200+ | Instant and professional | 81 ms |
| Inworld TTS-2 | $25 per 1M chars; $12.50 on Growth | $1.20; $0.60 | 200+ | Instant and professional | 166 ms |
| OpenAI tts-1 / tts-1-hd | $15 / $30 per 1M chars | $0.72 / $1.44 | Not compared | Not on the pricing page | Not on the board |
| Deepgram Aura-2 / Flux TTS | $30 / $45 per 1M chars | $1.44 / $2.16 | Aura-2: 7 | Not on the pricing page | 300 ms / 203 ms |
| Rime Mist v3 / Coda v2 | $0.030 / $0.050 per minute (20% off on Scale) | $1.80 / $3.00 | 4 / 10 | Enterprise only | 319 ms / 339 ms |
| Cartesia Sonic-3.6 | Plan credits, 1 per character: $5 for 100K to $299 for 8M | $2.40 to $1.79 | 44 | Instant on Pro; professional from $49 Startup | 292 ms |
| ElevenLabs Flash v2.5 / v4 Turbo | $40 per 1M chars | $1.92 | 32 / 90+ | Yes | 198 ms / 194 ms |
| ElevenLabs v4 | $80 per 1M chars | $3.84 | 90+ | Yes | Not on the board |
| MiniMax speech-2.8 Turbo / HD | $60 / $100 per 1M chars | $2.88 / $4.80 | 40 | Rapid cloning $1.50 a voice | Not on the board |
The latency column comes from Coval's time-to-first-audio benchmark, which times every model from sending text to the first audible sound. Coval sells voice-agent testing to many of these vendors. Vendor latency claims are lower (ElevenLabs states about 75 ms for Flash, Rime about 37 ms P50 for Mist v3 self-hosted) because most stop the clock at the first byte and exclude network time.
Which one to switch to#
For a real-time app or voice agent (the old PlayHT API and LMNT use)#
Inworld TTS-2 Flash is the cheapest per hour here and the fastest of these vendors in Coval's benchmark, with cloning and 200+ languages listed. Its weak spots: rates depend on your plan tier, and Inworld's speech-to-speech product has no per-minute price on its pricing page. Stork earns a commission on Inworld sign-ups through links here. ElevenLabs Flash or v4 Turbo costs about $1.92 an hour, more than twice Inworld Flash, but brings the most mature cloning and a full agent platform if you later want one; Stork also earns a commission there. Deepgram Aura-2 makes sense if you already use Deepgram for transcription, but it lists only seven languages and no cloning.
For voice cloning (the main reason many LMNT users chose it)#
ElevenLabs, Inworld and Cartesia all sell self-serve instant cloning. MiniMax charges $1.50 per cloned voice plus usage. Rime offers custom cloning only on Enterprise, and OpenAI and Deepgram list none. Whichever you pick, re-create voices from the original recordings: no vendor accepts another's voice IDs.
For predictable billing#
Rime bills per minute of audio rather than per character, has a free tier worth $50 of usage a month, and is built for call-centre voices. Test Coda on your own scripts first: Coval measured a 10.7% word error rate for it, against 1.5% to 3.1% for every other model on its board.
For the most languages#
Inworld lists 200+ languages and ElevenLabs v4 90+. Cartesia covers 44, MiniMax 40, Rime's Coda 10 and Deepgram's Aura-2 7. Check that the specific voice you want speaks your language: at Rime, for example, most Coda voices are English-only.
If you used Hume#
Hume's EVI was a speech-to-speech API, which needs a different kind of replacement. Our Hume AI alternatives guide covers EVI and Octave in detail, with a migration checklist.
Enjoying this? Get one like it in your inbox each morning.
one email a day · unsubscribe in two clicks · no third-party tracking
If you built on Vocode or Air AI#
- Vocode: the code is still on GitHub under an open-source licence, so a working deployment does not stop on its own, but nothing has been merged since November 2024. For new work, Pipecat and LiveKit Agents are the actively maintained open-source frameworks; Daily, which makes Pipecat, announced on October 29, 2026 that Deepgram is acquiring it and that Pipecat will stay open source and vendor neutral.
- Air AI: if anyone is still selling "Air AI" licences, read the FTC's settlement first. For an outbound or inbound AI calling platform, see our voice agent platforms guide.
How to avoid the next shutdown#
- Keep the original recordings for every cloned voice in your own storage.
- Call TTS through one function or an orchestrator (Pipecat, LiveKit) so the vendor is a setting, not a rewrite.
- Prefer vendors that publish dated changelogs and deprecation notices. Hume gave six weeks; PlayHT and LMNT gave customers no public dated notice we could find.
- Re-check prices at your real volume every quarter. Plan tiers change the rate by 10% (Deepgram) to over 50% (Inworld).
Frequently asked questions#
Is PlayHT still available?#
No. By January 6, 2026 the play.ht homepage read "We have shut down the service", and today the domain no longer loads. Meta acquired PlayAI, the company behind PlayHT, in July 2025.
What happened to LMNT?#
LMNT shut down. Its site now reads "LMNT has shut down. Our speech generation journey has come to an end." It gives no date, successor or refund terms.
What is the best PlayHT alternative for developers?#
For a real-time API, Inworld TTS-2 Flash is the cheapest per hour of audio we found and the fastest of these vendors in Coval's latency benchmark; ElevenLabs Flash or v4 Turbo costs more but has broader cloning and an agent platform. Both pay Stork a commission.
Which PlayHT alternative is cheapest?#
Inworld TTS-2 Flash and OpenAI tts-1 both list $15 per million characters on demand, about $0.72 per hour of audio at 800 characters a minute. Inworld's Growth plan lowers Flash to $7 per million.
Can I transfer my PlayHT or LMNT cloned voices?#
No vendor imports another's voice IDs, and both services are closed. If you still have the original recordings, you can re-create each voice on ElevenLabs, Inworld, Cartesia or MiniMax.
Is Vocode still maintained?#
Its main repository has had no commits since November 15, 2024 and vocode.dev now redirects to GitHub. The code remains available, but for new projects Pipecat and LiveKit Agents are actively maintained.