# 5 Coqui alternatives in 2026 for teams that built on XTTS-v2

By Samuel Oram-Jones

XTTS-v2, the Coqui model that clones a voice from a 6-second clip, ships under a licence that "allows only non-commercial use of a machine learning model and its outputs". The same licence adds that "Use for revenue-generating activity, including projects directly funded by government grants, is not a non-commercial purpose." A team that built a product on XTTS-v2 therefore has a licensing question for counsel to answer before it has a quality one, and the company that wrote the licence shut down in January 2024.

Five providers sell the same cloning work for paid products, as of November 2026: Bland Speech, ElevenLabs, Cartesia, Fish Audio, and Inworld.

## Why do teams look for Coqui alternatives?

Teams look for Coqui alternatives because Coqui shut down in January 2024, no hosted Coqui API or paid plan exists today, and XTTS-v2's licence allows only non-commercial use of the model and its outputs, so a product that earns revenue needs a provider that sells speech.

Coqui's own site said "Coqui is shutting down. Thank you for all your support!" in an Internet Archive copy from 14 January 2024. Coqui's former website address now returns a GitHub Pages 404 error, and Coqui's description of XTTS-v2 on Hugging Face still calls it "the same or similar model to what powers Coqui Studio and Coqui API", both of which are gone.

What survives is code and weights. The coqui-ai/TTS repository on GitHub isn't archived, carries an MPL-2.0 code licence, and has 46.3k stars, but its last commit is dated 10 February 2024 and its last release, v0.22.0, is dated 12 December 2023. Idiap Research Institute maintains a fork that calls the original "unmaintained", published release v0.28.0 on 19 October 2026, and installs with `uv pip install coqui-tts`. The XTTS-v2 weights themselves keep the Coqui Public Model License.

Coqui described XTTS-v2 as covering 17 languages, with cross-language cloning and 24 kHz output, and wrote that XTTS can stream with "<200ms latency" without naming the hardware.

## What should I check in a Coqui alternative?

A team replacing XTTS-v2 should check who grants the commercial rights, how much audio a clone needs, and whether the provider runs the GPUs, because XTTS-v2 answered the last question with a download and a non-commercial licence.

Commercial rights come from the provider's terms. ElevenLabs limits free workspaces to non-commercial use and licenses commercial use on paid plans, Cartesia's commercial use licence starts on Pro, and Inworld's On-Demand plan, at $0 a month, includes a commercial licence. Fish Audio's free plan is personal and non-commercial, and Fish Audio's open-source release comes under a community licence that makes commercial use free only while annual revenue is under USD 500,000.

Clone audio ranges from 3 seconds on Inworld to about 10 seconds on Bland Speech, ElevenLabs, and Cartesia, against 6 seconds for XTTS-v2. A professional clone, trained on about 30 minutes or more of recordings on Bland Speech, ElevenLabs, and Cartesia, adds a plan tier or a review step that a download never required.

Running GPUs is the work a hosted provider takes over. A team that moves from self-hosted XTTS-v2 to a hosted API trades GPU upkeep and a licence question for a per-character price.

## What are the best Coqui alternatives?

The best Coqui alternatives are Bland Speech, ElevenLabs, Cartesia, Fish Audio, and Inworld, with Bland Speech first because it clones from one clean sample of about 10 seconds on a hosted API, bills one standard per-character rate with no monthly plan, and ranked first of 13 AI voices in Design Arena's Audio Realism Bench results of 4 August 2026.

| Provider | How it is run | Instant clone audio | Published price |
| --- | --- | --- | --- |
| Bland Speech | Hosted API and studio | One clean sample of about 10 seconds | $0.04 per 1,000 characters, pay as you go |
| ElevenLabs | Hosted API | 10 seconds on Eleven v4 (Starter, $6 a month) | $0.08 per 1,000 characters on Eleven v4; $0.04 on Eleven v4 Turbo |
| Cartesia | Hosted API | 10 seconds (Pro, $5 a month) | Credits at 1 per character; Startup $49 a month for 1.25M |
| Fish Audio | Hosted API, plus open S2 weights | At least 10 seconds per clip | $15 per million UTF-8 bytes on s2.1-pro |
| Inworld | Hosted API | 3 seconds, on every plan | $25 per million characters on Realtime TTS-2 |
| XTTS-v2 (for comparison) | Self-run weights, non-commercial licence | 6 seconds | Free to download, with no hosted API |

<details><summary>What these rows mean</summary>

- **How it is run:** whether a provider hosts the model or hands over weights to run.
- **Instant clone audio:** the recording length for a clone made in minutes, with the plan that gates it where the provider names one.
- **Published price:** each provider's own rate with the plan or model named. Fish Audio bills per UTF-8 byte, which equals one character for plain English text.

</details>

### 1. Bland Speech

Bland Speech is Bland's own text to speech model, built for phone calls and sold through a studio and an API with HTTP and WebSocket streaming. An instant clone takes one clean sample of about 10 seconds and is ready in a minute. A professional clone is fine-tuned on 30 minutes or more of audio and goes through Bland's consent and rights review, and every clone is private to the customer's organization, with the customer confirming the voice is theirs to use. Output is PCM at 8 to 48 kHz or 8 kHz mu-law, and Bland publishes 199ms to first audio. The [text to speech API](/text-to-speech-api) has the endpoints.

Bland Speech suits a team that wants hosted speech with a consent step built into cloning. It costs $0.04 per 1,000 characters, or $40 per million, pay as you go with no monthly plan, and a new account gets 50,000 free characters once.

<div class="cmp-proscons cmp-proscons-own"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>An instant clone from one clean sample of about 10 seconds</li><li>Every clone private to the customer's organization, with the customer confirming each voice is theirs</li><li>Mu-law output at 8 kHz for phone lines</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>More than a hobby project needs, because Bland Speech is built for production calls</li></ul></div></div>

### 2. ElevenLabs

ElevenLabs sells Eleven v4 and Eleven v4 Turbo, both launched on 28 September 2026, and bills its API at $0.08 per 1,000 characters on v4 and $0.04 on v4 Turbo. An instant clone needs 10 seconds of audio on Eleven v4 and a Starter plan at $6 a month. A professional clone needs about 30 minutes and a Creator plan at $22 a month, and ElevenLabs verifies a professional clone with a voice captcha, where the voice owner reads lines aloud. ElevenLabs suits a team that wants 90+ languages and a large model family, and will manage plan tiers to get cloning.

<div class="cmp-proscons" data-rival="elevenlabs"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>90+ languages on the Eleven v4 family</li><li>Voice-captcha verification for professional clones</li><li>Mu-law and A-law output at 8 kHz</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>$0.08 per 1,000 characters on Eleven v4</li><li>Free workspaces are limited to non-commercial use</li><li>Professional clones need a Creator plan or above</li></ul></div></div>

### 3. Cartesia

Cartesia sells Sonic 3.6, released on 27 August 2026 in 44 languages, and says first audio arrives in under 90ms. An instant clone needs 10 seconds on the Pro plan at $5 a month, and up to 60 seconds helps a clone keep the speaker's accent. A professional clone needs at least 30 minutes, trains in up to 3 hours at no charge, and starts on Startup at $49 a month with 2 slots. Cartesia suits a team that wants a trained clone with free training and will manage a credit allowance, at 1 credit per character.

<div class="cmp-proscons" data-rival="cartesia"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>Professional clone training is free</li><li>44 languages on Sonic 3.6</li><li>On-prem and VPC deployment through Cartesia's sales team</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>Speech from a clone costs 1 credit per character against a monthly allowance</li><li>A professional clone starts on a $49 plan with 2 slots</li><li>Speed and volume are fixed on a professional clone</li></ul></div></div>

### 4. Fish Audio

Fish Audio is a voice AI platform with a web app, a REST API, and SDKs, and it open-sourced its S2 model on 9 March 2026. An instant clone needs at least 10 seconds per clip, and a minute or two of clear single-speaker speech improves fidelity. A professional clone runs in the web app only, needs a paid plan and at least 10 minutes of audio, and takes about 1 to 2 hours end to end. Fish Audio's API bills s2.1-pro at $15 per million UTF-8 bytes. Fish Audio suits a team that wants an open-weight model it can run itself, or a lower per-byte rate on a hosted clone.

<div class="cmp-proscons" data-rival="fish-audio"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>Open S2 weights, released on 9 March 2026</li><li>A free API model for development at $0</li><li>83 languages on s2.1-pro</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>The community licence makes commercial use free only while annual revenue is under USD 500,000</li><li>Professional cloning runs in the web app, not through the API</li><li>Output formats are mp3, wav, pcm, and opus</li></ul></div></div>

### 5. Inworld

Inworld sells realtime voice AI aimed at consumer apps, with Realtime TTS-2 at $25 per million characters on its On-Demand plan, which costs $0 a month and includes a commercial licence. Instant cloning works from as little as 3 seconds of audio, trimmed at a 30-second cap, and is included on every plan, while professional cloning is in beta, runs on TTS-2 only, and takes at least 10 minutes of audio. The clone form asks the user to confirm rights to the voice. Inworld suits a team that wants the shortest clone sample and plain-language direction of delivery.

<div class="cmp-proscons" data-rival="inworld"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>Instant cloning from 3 seconds on every plan</li><li>Cloned voices bill at the same rate as stock voices</li><li>Mu-law and A-law output at 8 kHz</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>Professional cloning is a beta that runs on TTS-2 only</li><li>Zero data retention leaves clone samples out</li></ul></div></div>

## Which Coqui alternative sounds most human?

Among the alternatives that were scored, Bland Speech v3 sounded most human, ranking first of 13 AI voices at 1365 in Design Arena's [Audio Realism Bench results of 4 August 2026](https://notes.designarena.ai/audio-realism-benchmark-measuring-realism-in-audio-models/), while a human recording scored 1500.

In those results Cartesia's Sonic 3.5 ranked 6th at 1096, ElevenLabs' Eleven v3 ranked 12th at 938, and Inworld's TTS-1.5 Max ranked 13th at 763. Coqui's XTTS-v2 and Fish Audio weren't scored, so those results don't rank either against Bland Speech v3. Eleven v4 (28 September 2026), Cartesia's Sonic 3.6 (27 August 2026), and Fish Audio's Drama 3 (13 October 2026) were released after those results and aren't scored in them. Design Arena scored one female and one male voice per model, each chosen as its most naturally conversational American-English voice, so a clone's sound depends on the sample behind it.

## Which Coqui alternative should I pick?

For a team leaving XTTS-v2, every provider on this list licenses commercial use, so the move comes down to how a clone is made and how it sounds on a call. Bland Speech makes a hosted clone from one clean sample of about 10 seconds, with a consent step on every voice and 8 kHz mu-law audio for phone calls, and Bland Speech v3 ranked first of 13 AI voices in Design Arena's results of 4 August 2026.

A team that must keep running its own weights has open options outside this list. Kokoro is an 82-million-parameter model under an Apache 2.0 licence, trained by an individual author, with 54 voices across 8 languages in its v1.0 release. Orpheus, from Canopy Labs, is an Apache 2.0 model on a Llama 3.2 3B base whose public repository has had no commits since 5 December 2025, and Canopy Labs still has a fix to its voice cloning notebook open on its checklist. [Open-source text to speech compared](/blog/open-source-tts-compared) covers those models.

Price doesn't make the case for Bland Speech. At $40 per million characters it costs more than Fish Audio's $15 per million UTF-8 bytes and Inworld's $25, the same as ElevenLabs' $40 for Eleven v4 Turbo, and less than ElevenLabs' $80 for Eleven v4, and a self-run XTTS-v2 costs nothing to download, which is why its licence matters. The case for Bland Speech rests on the realism result and phone-ready audio.

[Bland Speech vs Fish Audio](/compare/bland-speech-vs-fish-audio) and [Bland Speech vs ElevenLabs voice cloning](/compare/bland-speech-vs-elevenlabs-voice-cloning) go deeper on two of these providers, and [How to move a cloned voice to Bland Speech](/blog/move-cloned-voice-to-bland-speech) covers the steps for a voice cloned elsewhere. The [alternatives guides](/compare/text-to-speech/alternatives) cover other providers, the [text to speech hub](/compare/text-to-speech) holds every comparison, and [Bland Speech](/speech) has a studio for a first clone.

## Frequently asked questions

### What is the best Coqui alternative?

Bland Speech is the best Coqui alternative for phone calls and voice agents, because it clones from one clean sample of about 10 seconds, bills $0.04 per 1,000 characters with no monthly plan, and ranked first of 13 AI voices in Design Arena's results of 4 August 2026.

### Did Coqui shut down?

Yes, Coqui shut down in January 2024. An Internet Archive copy of Coqui's site from 14 January 2024 reads "Coqui is shutting down. Thank you for all your support!", and Coqui's former website address now returns a GitHub Pages 404 error.

### Can XTTS-v2 be used in a paid product?

Not under its own licence, which "allows only non-commercial use" and says revenue-generating activity is not a non-commercial purpose. A team with a revenue-bearing product takes the question to counsel, or moves to a provider that sells speech under commercial terms.

### Is the Coqui TTS code still maintained?

The original coqui-ai/TTS repository has had no commit since 10 February 2024. Idiap Research Institute's fork published release v0.28.0 on 19 October 2026 and calls the original unmaintained.

### Can a Coqui voice move to another provider?

No, a cloned voice stays with the system that trained it, so a switch means cloning again from the original recordings, with the speaker's consent. [How to move a cloned voice to Bland Speech](/blog/move-cloned-voice-to-bland-speech) covers the steps for Bland.

### How much audio does a clone need on each alternative?

An instant clone needs 3 seconds on Inworld, about 10 seconds on Bland Speech, 10 seconds on ElevenLabs and Cartesia, and at least 10 seconds per clip on Fish Audio, against 6 seconds for XTTS-v2. A professional clone needs at least 10 minutes of audio on Fish Audio and Inworld, 30 minutes or more on Bland Speech and Cartesia, and about 30 minutes on ElevenLabs.
