# Inworld vs MiniMax (2026): Which text to speech API is worth the price?

By Samuel Oram-Jones

A benchmark score only helps a buyer when it belongs to the model that buyer will call, and for both companies here it doesn't. In Design Arena's Audio Realism Bench results of 4 August 2026, MiniMax's Speech-02 HD ranked 7th of 13 AI voices at 1093 and Inworld's TTS-1.5 Max ranked 13th at 763, a gap of 330 Elo points, but MiniMax's current Speech 2.8 and Inworld's Realtime TTS-2 and TTS-2 Flash weren't in those results.

What can be compared today is the price list and the feature sheet. As of November 2026, Inworld charges $25 per million characters for TTS-2 and $15 for TTS-2 Flash, and MiniMax charges $100 for speech-2.8-hd and $60 for speech-2.8-turbo. [Bland Speech](/speech) v3, which ranked first in those 4 August 2026 results at 1365, is the third option a phone team weighs here.

## Which is better, Inworld or MiniMax?

Inworld is the better fit for a team watching cost, and MiniMax is the better fit for a team that wants inline performance tags and an emotion setting. Inworld's flagship is a quarter of MiniMax's price at $25 against $100 per million characters, and Inworld documents Pipecat, LiveKit, Vapi, and Twilio ConversationRelay routes, plus an OpenAI-compatible endpoint that shipped on 11 September 2026.

| Feature | Bland Speech | Inworld | MiniMax |
|---|---|---|---|
| Audio Realism Bench, 4 August 2026 | 1st at 1365 (Bland Speech v3) | 13th at 763 (TTS-1.5 Max) | 7th at 1093 (Speech-02 HD) |
| Price per million characters | $40, one standard rate, pay as you go | TTS-2 $25, TTS-2 Flash $15 on pay as you go, down to $12.50 and $7 on Growth | speech-2.8-hd $100, speech-2.8-turbo $60 |
| Monthly plan | None | $0 on-demand, then $25 to $1,500 a month, or Enterprise | Optional audio packages, $5 to $999 a month |
| First audio, as each company publishes it | 199ms | 20ms to 100ms, depending on the measure | "Ultra-low latency" (speech-2.6-hd) |
| Instant cloning | One clean sample of about 10 seconds | As little as 3 seconds, on every plan | 10 seconds to 5 minutes, $1.50 a voice |
| Phone audio | Mu-law at 8 kHz | Mu-law and A-law at 8 kHz | Mu-law at 8 kHz |
| Text streamed in from an LLM | WebSocket | WebSocket with sentence-boundary mode | Bidirectional WebSocket |

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

- **Audio Realism Bench:** Design Arena's blind listening test, in which listeners hear two clips of one line and pick the one that sounds more like a real person. A higher Elo rating means a voice was picked more often. The scored models are named in the cells, and each company now sells newer ones.
- **Price per million characters:** Inworld's On-Demand plan is $0 a month and bills each character as used, and its paid plans lower the rate as the monthly fee rises. MiniMax's figures are its text to speech rates on pay as you go.
- **First audio:** The time from sending text to hearing the first sound. Inworld gives 100ms for TTS-2 and 20ms for TTS-2 Flash at the 90th percentile, measured at the server without network time, and 90ms and as low as 50ms on a third-party benchmark. MiniMax calls speech-2.6-hd "Ultra-low latency". Bland publishes 199ms.
- **Phone audio:** Mu-law and A-law are the two compressed formats phone networks use, mu-law in North America.

</details>

## Where does Inworld fall short next to MiniMax?

Inworld falls short on its realism score, on a short REST limit, and on its own language counts. TTS-1.5 Max ranked 13th at 763 in the 4 August 2026 results, a REST synthesis call takes up to 2,000 characters where MiniMax's HTTP call takes up to 10,000, and Inworld gives three language counts: 500+ languages and dialects, 200+ languages and locales, and 470 languages across 623 codes.

Latency numbers vary by source as well. Inworld says TTS-2 Flash delivers first audio in as low as 50ms and TTS-2 in as low as 90ms at the 90th percentile on Coval, a third-party benchmark, while Inworld's own figures are 20ms and 100ms measured at the server without network time, and Inworld cites a Cekura benchmark with medians of 45ms and 127ms.

Inworld's strengths are the price ladder and the controls. Pay as you go is $25 and $15 per million characters, and Creator, Builder, Developer, and Growth plans bring the rate to $20, $17.50, $15, and $12.50 for TTS-2, with Enterprise as low as $5. TTS-2 takes direction in plain words, either as a request-level instruction or inline in square brackets, though steering works on TTS-2 only and TTS-2 Flash ignores it. Both models render 50 non-verbal sounds such as [laugh] and [sigh], and SSML break tags add pauses.

### What Inworld costs

On-Demand is $0 a month and includes up to 70 minutes of text to speech, 100 custom voices, 2 professional voice clones, and a commercial license. Creator is $25 a month, Builder $100, Developer $300, and Growth $1,500, each with an equal amount in credits that roll over for 3 months. Cloned and designed voices bill at the same rate as stock voices. An instant clone needs as little as 3 seconds of audio, up to a 30-second cap, and the clone form asks the user to confirm the rights to the voice. Professional cloning is in beta, needs at least 10 minutes of audio, trains in 30 to 60 minutes, and works on TTS-2 only. Inworld says zero data retention is on Developer, Growth, and Enterprise, and that clone samples are not covered by it.

Inworld says it is SOC 2 Type II certified and HIPAA and GDPR compliant, with a BAA, a HIPAA business associate agreement, on Enterprise or as a Growth add-on. Enterprise customers can keep data in the EU or India. Inworld describes its text to speech as realtime voice AI for consumer-facing applications.

<div class="cmp-proscons" data-rival="inworld"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>$25 and $15 per million characters on pay as you go, lower on paid plans</li><li>Plain-word steering on TTS-2 and 50 non-verbal sounds on both models</li><li>Instant cloning from 3 seconds of audio on every plan</li><li>Pipecat, LiveKit, Vapi, and an OpenAI-compatible endpoint</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>TTS-1.5 Max ranked 13th at 763 in the 4 August 2026 results</li><li>REST calls take up to 2,000 characters</li><li>Steering works on TTS-2 only</li><li>Zero data retention doesn't cover clone samples</li></ul></div></div>

Inworld is a poor fit for:

- A script that has to go to a REST endpoint in one piece above 2,000 characters
- A buyer who needs a single language count to plan a rollout

## Where does MiniMax fall short next to Inworld?

MiniMax falls short on price, because speech-2.8-hd at $100 per million characters is four times Inworld's TTS-2 at $25, and speech-2.8-turbo at $60 is four times TTS-2 Flash at $15. MiniMax also sells monthly audio packages in audio points, from $5 a month for 100,000 points to $999 for 20 million, with yearly billing 20% off, and the packages cover every speech model.

MiniMax's cloning has a clock on it. Rapid cloning takes 10 seconds to 5 minutes of audio at $1.50 a voice, charged when the clone is first used, and MiniMax says "If a cloned voice is not used within 7 days, the system will delete it." Interjection tags work on speech-2.8 models only, and the 2.8 models don't support the whisper emotion.

On control, MiniMax holds its own. Speech-2.8 models take 19 inline interjection tags, among them laughs, sighs, and breath, an emotion setting with seven values on 2.8, pause markers from 0.01 to 99.99 seconds, and pronunciation overrides in Pinyin, IPA, or Jyutping. A voice can be designed from a text description for $3. MiniMax's bidirectional WebSocket accepts LLM output at any granularity, even character by character, a `task_cancel` message interrupts a turn, and async jobs run to 1 million characters per request. MiniMax runs 40 languages on its 2.6 and 2.8 models, and describes itself as a general AI company with speech as one product line.

<div class="cmp-proscons" data-rival="minimax"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>19 inline interjection tags and seven emotion values on speech-2.8</li><li>Bidirectional WebSocket with cancel, and async jobs to 1 million characters</li><li>HTTP requests up to 10,000 characters</li><li>40 languages on the 2.6 and 2.8 models</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>$100 and $60 per million characters, four times Inworld's flagship rates</li><li>A cloned voice is deleted after 7 days without use</li><li>Interjection tags work on speech-2.8 models only</li></ul></div></div>

MiniMax is a poor fit for:

- A cost target near $25 per million characters
- A cloned voice that sits idle for a week between campaigns

## Which sounds more human, Inworld or MiniMax?

MiniMax's scored model ranked well above Inworld's in the results that scored both, though neither company's current model was in them. In Design Arena's Audio Realism Bench results of 4 August 2026, MiniMax's Speech-02 HD ranked 7th of 13 AI voices at 1093, and Inworld's TTS-1.5 Max ranked 13th at 763.

Speech-02 dates from April 2025, and MiniMax released Speech 2.8 on 23 January 2026. Inworld launched TTS 1.5 on 21 January 2026, Realtime TTS-2 on 5 May 2026, and TTS-2 Flash on 9 August 2026, after those results. Bland Speech v3 ranked first in those results at 1365, human recordings scored 1500, and Design Arena explains [how the listening test works](https://notes.designarena.ai/audio-realism-benchmark-measuring-realism-in-audio-models/), including that every comparison uses two voices of the same gender.

## Why do people comparing Inworld and MiniMax also weigh Bland Speech?

People comparing Inworld and MiniMax also weigh Bland Speech because Bland Speech's first place at 1365 on 4 August 2026 belongs to its current model, and neither rival's score does. Bland Speech bills $0.04 per 1,000 characters ($40 per million), pay as you go with no monthly plan, and 50,000 free characters once for a new account. That is above Inworld's $25 and $15 and below MiniMax's $60 and $100.

Bland publishes 199ms to first audio, with each company's figure as it publishes it. Bland Speech returns mu-law at 8 kHz or PCM from 8 to 48 kHz, streams over HTTP and a WebSocket where a new context interrupts the turn that is playing, and has first-party Pipecat and LiveKit Agents integrations, plus an OpenAI-compatible endpoint. An instant clone needs one clean sample of about 10 seconds, and a professional clone is fine-tuned on 30 minutes or more of audio and goes through Bland's consent and rights review.

Bland runs its own model and speech, and Bland is SOC 2 Type II audited, HIPAA compliant with a signed BAA, and FedRAMP 20x Class C certified at Moderate, granted 17 November 2026, with speech synthesis inside that boundary. Enterprise pricing is custom, set with Bland's sales team.

<div class="cmp-proscons cmp-proscons-own"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>Ranked first in Design Arena's Audio Realism Bench results of 4 August 2026, at 1365</li><li>One standard rate, $0.04 per 1,000 characters, with no monthly plan</li><li>Mu-law at 8 kHz and a WebSocket with barge-in</li><li>An OpenAI-compatible endpoint and Pipecat and LiveKit integrations</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>Built around live phone calls, so a team that only wants a low-cost batch voice may not need it</li></ul></div></div>

For phone calls and voice agents, Bland Speech is the pick over both, because its first-place score belongs to the model a team would actually call. A team can read [Bland Speech vs Inworld](/compare/bland-speech-vs-inworld), [Bland Speech vs MiniMax](/compare/bland-speech-vs-minimax), and the [Inworld TTS review](/blog/inworld-tts-review) for each rival alone, and the [text to speech hub](/compare/text-to-speech) and [rival vs rival pairings](/compare/text-to-speech/rival-vs-rival) for more. Endpoints and sample code sit on the [text to speech API](/text-to-speech-api) page.

## Frequently asked questions

### Is Inworld better than MiniMax?

Inworld is better on price, and MiniMax is better on the 4 August 2026 score. Inworld's TTS-2 is $25 per million characters against $100 for MiniMax's speech-2.8-hd, while MiniMax's Speech-02 HD ranked 7th at 1093 and Inworld's TTS-1.5 Max ranked 13th at 763, both older models.

### Which is cheaper, Inworld or MiniMax?

Inworld is cheaper. TTS-2 is $25 per million characters and TTS-2 Flash is $15, against $100 for MiniMax's speech-2.8-hd and $60 for speech-2.8-turbo, and Inworld's paid plans lower the rates to $12.50 and $7 on Growth. Bland Speech is $40 per million characters.

### Which has lower latency, Inworld or MiniMax?

Inworld publishes figures of 20ms to 100ms, measured at the server without network time, and 50ms to 90ms on a third-party benchmark. MiniMax calls speech-2.6-hd "Ultra-low latency". Bland publishes 199ms for Bland Speech, with each figure as that company publishes it.

### Can Inworld and MiniMax clone a voice?

Both can. Inworld clones from as little as 3 seconds of audio on every plan and offers professional cloning in beta from 10 minutes of audio. MiniMax clones from 10 seconds to 5 minutes at $1.50 a voice and deletes a clone left unused for 7 days.

### Do Inworld and MiniMax work on phone calls?

Both return mu-law audio at 8 kHz, the compressed format North American phone networks use. Inworld also offers A-law, publishes a Twilio ConversationRelay guide, and has Pipecat, LiveKit, and Vapi integrations.

### Which sounds more human, Inworld or MiniMax?

MiniMax's scored model did. In Design Arena's Audio Realism Bench results of 4 August 2026, MiniMax's Speech-02 HD ranked 7th at 1093 and Inworld's TTS-1.5 Max ranked 13th at 763, with Bland Speech v3 first at 1365.

### What is a good alternative to Inworld and MiniMax?

Bland Speech is the alternative to weigh for a phone line. Bland Speech v3 ranked first in those 4 August 2026 results, bills $0.04 per 1,000 characters with no monthly plan, and returns mu-law at 8 kHz.
