# Bland Speech vs Qwen TTS (2026): Pay per character, or run Alibaba's open model on a team's own GPUs

A team that searches for Qwen TTS finds two different purchases under one name. Alibaba Cloud sells its Qwen-Audio models through Model Studio, billed by the character or the token, and the Qwen team publishes Qwen3-TTS, an open-weight model under the Apache 2.0 license that a team can run on its own hardware.

Bland Speech sits against both. Bland Speech v3 scored 1365 in Design Arena's blind listening test of 4 August 2026, and no Qwen model was in those results, so the comparison rests on price, phone audio, controls, cloning, and where each model runs.

## Which is better, Bland Speech or Qwen TTS?

For a voice agent on a phone line where callers should hear a person, Bland Speech is the better choice as of November 2026. Bland Speech v3 scored 1365 in Design Arena's Audio Realism Bench results of 4 August 2026, and it returns mu-law at 8 kHz, the format phone networks carry.

Qwen TTS makes more sense for a team that wants an open-weight model it can run itself, a lower price per character, free-form instructions for how a line is delivered, or Chinese dialects.

<div class="cmp-usewhen"><div class="cmp-usewhen-own"><p><strong>Use Bland Speech when</strong></p><ul><li>Callers on a phone line should hear a voice that sounds like a person</li><li>A team wants one pay as you go rate per character, with no token conversion</li><li>An existing OpenAI audio speech integration should switch providers by changing a base URL, a key, and a model name</li><li>A script needs performed delivery, such as a laugh or a warm read, written inline as tags</li><li>A voice agent runs on Pipecat or LiveKit Agents and needs a text to speech service built for those frameworks</li></ul></div><div><p><strong>Use Qwen TTS when</strong></p><ul><li>A team wants to download Qwen3-TTS under Apache 2.0 and run it on its own hardware</li><li>A team wants a lower price per character and will pay $10 per million characters on qwen3-tts-flash</li><li>A script needs free-style instructions for emotion, tone, persona, or dialect, or SSML on supported voices</li><li>A team already runs on Alibaba Cloud and will choose between the international and China regions</li></ul></div></div>

| Feature | Bland Speech | Qwen TTS |
| --- | --- | --- |
| Audio Realism Bench, 4 August 2026 | Scored 1365, ranked 1st of 13 AI voices | Not in those results |
| Price per million characters | $40, one standard rate | $10 on qwen3-tts-flash in the international scope, $20 or $30 on Qwen-Audio-3.1-TTS-Plus by region, per-token on Qwen-Audio-3.1-TTS-Flash |
| Open weights | A closed-source model Bland built itself, reached through an API | Qwen3-TTS, Apache 2.0, 0.6B and 1.7B sizes |
| First audio, as each company publishes it | 199ms | "As low as 97ms" for open-weight Qwen3-TTS |
| Phone audio | Mu-law at 8 kHz, PCM from 8 to 48 kHz | mp3, pcm, wav, and opus at 8 to 48 kHz |
| Delivery controls | Inline tags such as [laughs] and [say warmly], expressiveness and stability dials, timed pauses | Free-style instructions and tags for emotion, tone, persona, rate, and volume, SSML on supported voices |
| Instant voice clone | One clean sample of about 10 seconds | Five seconds or more, at $0.01 per enrollment call in the international scope |

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

- **Audio Realism Bench, 4 August 2026:** Audio Realism Bench is Design Arena's blind listening test, where native speakers hear two clips of the same line from two models and pick the one that sounds more like a real person. A higher score means listeners picked that voice more often.
- **Price per million characters:** Bland Speech costs $0.04 per 1,000 characters, which is $40 per million. Alibaba Cloud prices Qwen-Audio-3.1-TTS-Plus at $0.20 or $0.30 per 10,000 characters, depending on region, and qwen3-tts-flash at $0.10 per 10,000 characters in its international scope. Qwen-Audio-3.1-TTS-Flash bills input and output tokens, priced by region.
- **Open weights:** Weights are the trained model files. A team that downloads open weights runs the model itself, and Bland Speech is reached through Bland's API.
- **First audio:** Each is the company's own published claim. Bland publishes 199ms to first audio on a streamed request. The Qwen team claims synthesis latency as low as 97ms for the open-weight model.
- **Phone audio:** US phone networks carry mu-law at 8 kHz, which Bland Speech returns directly. Alibaba Cloud's formats are mp3, pcm, wav, and opus, with a sample rate of 8,000 Hz available.

</details>

## How do Bland Speech and Qwen TTS compare?

The difference is where each model lives. Bland Speech is a model built for phone calls and reached through an API, and Qwen is a family: a hosted service on Alibaba Cloud with its own model ids, and an open-weight release that a team can run itself.

### Bland Speech overview

Bland Speech is the text to speech model inside Bland, a voice AI provider that builds and runs its own language, speech, and transcription models for enterprise phone calls. Bland Speech is sold on its own as an API and a studio on the [Bland Speech](/speech) page.

A team sends text over HTTP to POST /v2/tts and gets raw audio or a WAV file back, or streams language model tokens into a WebSocket at WSS /v2/tts/ws, where a new context interrupts the turn that is playing. POST /v2/audio/speech follows the OpenAI audio speech format, and a first-party Pipecat service and a LiveKit Agents plugin for Python are available. The [text to speech API](/text-to-speech-api) page has the code.

<div class="cmp-proscons cmp-proscons-own"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>Bland Speech v3 ranked 1st of 13 AI voices on Audio Realism Bench on 4 August 2026, at 1365</li><li>Pay as you go at $0.04 per 1,000 characters, with no monthly plan</li><li>An OpenAI-compatible endpoint, so an existing integration moves over by changing a base URL</li><li>Performance tags, and a director in the studio for how a line is read</li><li>Mu-law at 8 kHz and PCM from 8 to 48 kHz, ready for phone calls</li><li>Instant clones from about 10 seconds of audio, and professional clones from 30 minutes or more</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>More than a quick prototype needs, because Bland Speech is built for production calls</li></ul></div></div>

### Qwen TTS overview

Alibaba Cloud Model Studio sells Qwen-Audio-3.1, in a Plus model for brand voices, audio content, film and television dubbing, and character performance, and a Flash model for real-time interactions such as voice assistants and customer service. Qwen-Audio-3.0 and the older qwen3-tts-flash models are still sold beside them. The Qwen team also released the open-weight Qwen3-TTS on 22 January 2026, in 0.6B and 1.7B sizes, with a base model that clones a voice from three seconds of audio.

On the hosted API, a WebSocket accepts text in sequence and returns audio after each complete sentence, and an HTTP endpoint covers non-streaming and server-sent events, with word-level timestamps on streaming output. Alibaba Cloud's HTTP API names language codes for Chinese, English, French, German, Japanese, Korean, Russian, Portuguese, Thai, Indonesian, Vietnamese, Spanish, Italian, Malay, Filipino, and Arabic, and a further group of about 50 languages for Qwen-Audio-3.1-TTS-Flash.

<div class="cmp-proscons" data-rival="qwen-tts"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>$10 per million characters on qwen3-tts-flash in Alibaba Cloud's international scope</li><li>Open-weight Qwen3-TTS under Apache 2.0</li><li>Free-style instructions for emotion, tone, persona, and dialect, plus SSML on supported voices</li><li>Word-level timestamps on streaming output</li><li>A free quota for new Singapore activations, valid for 90 days</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>Prices differ by region, and Qwen-Audio-3.1-TTS-Plus costs $0.20 or $0.30 per 10,000 characters depending on which one</li><li>Qwen-Audio-3.1-TTS-Flash bills tokens, so a character price needs a conversion</li><li>The WebSocket returns audio after each complete sentence of text it receives</li><li>A team that runs Qwen3-TTS provides its own hardware</li></ul></div></div>

### How do Bland Speech and Qwen TTS handle phone audio and latency?

Bland Speech and Qwen publish latency for different products. Bland publishes 199ms to first audio on a streamed request. The Qwen team says Qwen3-TTS, the open-weight model, reaches end-to-end synthesis latency as low as 97ms and can output a first audio packet after a single character of input. Alibaba Cloud calls qwen3-tts-flash low-latency and describes Qwen-Audio-3.1-TTS-Flash as a model for real-time interactions.

Bland Speech outputs mu-law at 8 kHz and PCM at 8, 16, 24, 44.1, or 48 kHz. Alibaba Cloud outputs mp3, pcm, wav, and opus, at sample rates of 8,000, 12,000, 16,000, 22,050, 24,000, 44,100, and 48,000 Hz, with opus limited to 8,000, 12,000, 16,000, 24,000, and 48,000.

### How do Bland Speech and Qwen TTS handle security?

Alibaba Cloud's compliance programmes across its cloud include ISO 27001, SOC 2 Type 2, PCI DSS, and HIPAA (United States). Model Studio is sold in an international region based in Singapore and a China (Beijing) region, and the Qwen3-TTS weights can run on hardware a team controls.

Bland is SOC 2 Type II audited, HIPAA compliant with a signed BAA, and FedRAMP 20x Class C certified at Moderate, granted on 17 November 2026. Speech synthesis runs inside that FedRAMP boundary, and Bland runs its own model and speech instead of wrapping a third party, as [Bland's trust and security record](/trust-security) sets out. Enterprise customers get dedicated orchestration and dedicated GPUs run by Bland.

## How much do Bland Speech and Qwen TTS cost?

Qwen costs less per character than Bland Speech on every hosted model priced by the character, from $10 to $30 per million depending on model and region, against Bland's $40, and the open-weight model has no per-character fee at all.

### Bland Speech pricing

Bland Speech costs $0.04 per 1,000 characters, directed or not, in the studio or the API. New accounts get 50,000 characters free, and credits do not expire. A 60-second note costs about $0.04, a 10-minute narration about $0.36, and a million characters $40.

Enterprise pricing is custom, set with Bland's sales team.

### Qwen TTS pricing

Alibaba Cloud sets list prices by region. Qwen-Audio-3.1-TTS-Plus costs $0.20 or $0.30 per 10,000 characters, depending on region. In Alibaba Cloud's international scope, Qwen-Audio-3.0-TTS-Plus costs $0.20 per 10,000 characters, qwen3-tts-flash $0.10, and qwen3-tts-flash-realtime $0.13. Qwen-Audio-3.1-TTS-Flash bills input and output tokens instead, at prices that also differ by region. Voice enrollment costs $0.01 per call in the international scope.

New Model Studio activations in Singapore get a free quota valid for 90 days, and the quota applies in Singapore only.

### What does a month of speech cost on each?

At 100,000 characters a month, about two hours of speech, Bland Speech costs $4, qwen3-tts-flash costs $1, and Qwen-Audio-3.1-TTS-Plus costs $2 or $3 by region. At 1 million characters, about 20 hours, the figures are $40, $10, and $20 or $30. At 10 million characters, about 200 hours, they are $400, $100, and $200 or $300, with qwen3-tts-flash at its international list price.

Running Qwen3-TTS from the open weights moves the cost to hardware and operations, which depends on the GPUs a team already owns.

## Which sounds more human, Bland Speech or Qwen TTS?

Design Arena's blind tests of 4 August 2026 scored Bland Speech v3 at 1365, second only to a recording of a real person, and no Qwen model was in them. Qwen3-TTS, the open-weight model, came out on 22 January 2026.

Design Arena published how its listeners heard two clips of the same line without knowing which model made which, in [an article on Audio Realism Bench](https://notes.designarena.ai/audio-realism-benchmark-measuring-realism-in-audio-models/). The article's prompt categories were phone agents, explainers, and conversations, and acting-heavy speech was out of scope.

## Why is Bland Speech the best Qwen TTS alternative?

A team that wants one voice to call from a phone line, with one character price and no token billing to convert, is the buyer Bland Speech fits, and Bland Speech v3 scored 1365 in the 4 August 2026 tests.

Bland Speech bills characters at one standard rate, so a forecast is the characters a team speaks multiplied by $0.04 per 1,000, and an existing OpenAI audio speech integration can switch by changing a base URL, a key, and a model name. A brand voice clones from about 10 seconds of audio, or from 30 minutes or more after Bland's consent and rights review, and every clone stays private to the organization that made it.

Bland's [comparison of text to speech APIs](/compare/text-to-speech) covers more vendors, and the [head-to-head comparisons](/compare/text-to-speech/head-to-heads) collect the pairs. A team weighing open weights against an API can read [open-source text to speech vs a hosted API](/blog/open-source-tts-vs-hosted-api), and one comparing price first can start with the [text to speech API pricing comparison](/blog/text-to-speech-api-pricing).

## Frequently asked questions

### Is Bland Speech better than Qwen TTS?

For phone calls and voice agents where the voice should sound like a person, Bland Speech is better than Qwen TTS. Bland Speech v3 scored 1365 on Audio Realism Bench on 4 August 2026, and no Qwen model was in those results. Qwen suits a team that wants open weights or a lower price per character.

### Is Qwen TTS free?

The open-weight Qwen3-TTS is free to download and run under Apache 2.0, and the hosted Qwen models are priced. Alibaba Cloud charges from $0.10 per 10,000 characters on qwen3-tts-flash in its international scope, and new Singapore activations get a free quota for 90 days.

### How much does Qwen TTS cost?

In Alibaba Cloud's international scope, qwen3-tts-flash costs $10 per million characters, and Qwen-Audio-3.1-TTS-Plus costs $20 or $30 per million depending on region. Qwen-Audio-3.1-TTS-Flash bills by the token. Bland Speech costs $40 per million characters.

### Is Qwen TTS open source?

The Qwen3-TTS weights are open, released by the Qwen team under Apache 2.0 in 0.6B and 1.7B sizes. The hosted Qwen-Audio-3.0 and 3.1 models are sold through Alibaba Cloud Model Studio.

### Does Qwen TTS clone voices?

Yes. Alibaba Cloud's voice enrollment copies a voice from five seconds or more of audio, and the open-weight base model clones from a three-second sample. Alibaba Cloud offers an optional invisible AIGC identifier in WAV, MP3, or Opus audio, off by default.

### Which is faster, Bland Speech or Qwen TTS?

Each company publishes a different claim. Bland publishes 199ms to first audio on a streamed request, and the Qwen team says open-weight Qwen3-TTS reaches latency as low as 97ms, with each figure coming from its own company's method.

### Does Bland Speech work on phone calls?

Yes. Bland built Bland Speech for phone calls, and it returns mu-law at 8 kHz as well as PCM from 8 to 48 kHz, so audio goes onto a phone line without conversion.
