# 5 Inworld TTS alternatives in 2026 for live voice agents

By Samuel Oram-Jones

Inworld publishes three different first-audio figures for one model. Inworld says TTS-2 Flash reaches P90 time to first audio byte in 20ms, measured server-side and excluding network time, cites a Cekura benchmark with a 45ms median, and advertises first audio "as low as 50 ms" measured on Coval. A team building a voice agent has to plan around a single number, and picking among those three is the first reason to price out other text to speech APIs, as of November 2026.

Five options are worth pricing: Bland Speech, Cartesia, ElevenLabs, Murf, and Google Cloud. Bland Speech comes first because Bland Speech v3 ranked first of 13 AI voices in Design Arena's [Audio Realism Bench results of 4 August 2026](https://notes.designarena.ai/audio-realism-benchmark-measuring-realism-in-audio-models/), and because it returns phone-ready audio at one published standard rate. The [text to speech hub](/compare/text-to-speech) and the [alternatives family page](/compare/text-to-speech/alternatives) cover the wider market, and [Bland Speech vs Inworld](/compare/bland-speech-vs-inworld) and an [Inworld TTS review](/blog/inworld-tts-review) go deeper on Inworld itself.

## Why do teams look for Inworld TTS alternatives?

Teams look for Inworld TTS alternatives when claims on Inworld's own pages disagree, when the fastest model can't be steered, or when a listening test puts the voice behind its peers, and each of those has a published source.

Inworld sells realtime voice AI for consumer-facing applications, and its flagship is Realtime TTS-2, launched on 5 May 2026, with TTS-2 Flash, the fastest and cheapest model, following on 9 August 2026. It streams over HTTP and a WebSocket that takes text as a language model writes it, returns mu-law and A-law at 8 kHz, and ships an OpenAI-compatible endpoint, a Twilio ConversationRelay guide, and plugins for Pipecat and LiveKit. On-Demand costs $0 a month with TTS-2 at $25 per million characters and Flash at $15, and Enterprise goes as low as $5.

Three catches sit in Inworld's published material. Inworld says "500+ languages and dialects" in one place and "200+ languages and locales" in another, so a buyer comparing language coverage has two figures to choose from. Steering, where a developer describes the delivery in plain words, works on TTS-2 only and Flash ignores it, which means a team picks between the fastest model and the most directable one. Zero data retention starts on the $300 Developer plan and doesn't cover voice cloning samples.

A fourth comes from a listening test. Inworld's TTS-1.5 Max ranked 13th of 13 AI voices at 763 in Design Arena's results of 4 August 2026, and the newer TTS-2 and TTS-2 Flash weren't scored in them, so the result describes an older model.

## What should I check in an Inworld TTS alternative?

A buyer should check which model a latency figure belongs to, where it was measured, which features that model supports, and what the plan adds, because the same company can publish several figures and several feature sets.

Latency comes first. A server-side P90 (the time within which 90% of requests return) that excludes the network answers a different question from a median on a benchmark, and neither equals what a caller hears on a phone line. Cartesia says under 90ms, ElevenLabs says a median of about 100ms for Eleven v4 Turbo excluding application and network time, Murf says under 100ms for Falcon 2.1, and Bland publishes 199ms to first audio on a streamed request.

Features come second, because tags, steering, and cloning attach to models and plans: ElevenLabs' v3 and v4 take audio tags but not SSML break tags, Cartesia's emotion controls are beta and English only, and Bland's performance tags work on current-generation voices only.

Phone audio comes third. Mu-law is the compressed 8 kHz format phone networks carry, and Inworld, Cartesia, ElevenLabs, Murf, Google Cloud, and Bland Speech all offer it.

## What are the best Inworld TTS alternatives?

The best Inworld TTS alternatives are Bland Speech, Cartesia, ElevenLabs, Murf, and Google Cloud, with Bland Speech first for phone calls and voice agents because of its realism result and its published price and latency.

| Alternative | Best for | Published price | Audio Realism Bench, 4 August 2026 |
| --- | --- | --- | --- |
| Bland Speech | Phone calls and voice agents | $40 per million characters, pay as you go | Bland Speech v3: ranked 1st, 1365 |
| Cartesia | A sub-90ms claim across 44 languages | 1 credit per character, plans from $5 a month | Sonic 3.5: ranked 6th, 1096 |
| ElevenLabs | A large voice library across 90+ languages | $40 to $80 per million characters, by model | Eleven v3: ranked 12th, 938 |
| Murf | A low per-character rate | $10 per million characters for Falcon 2.1, $30 for Gen2 | Wasn't in those results |
| Google Cloud | Teams already on Google Cloud | $30 per million characters for Chirp 3: HD, first million free each month | Gemini 2.5 Pro TTS: ranked 5th, 1109 |

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

- **Published price:** each vendor's own rate with the plan or model named. Cartesia sells credits at one per character, with overages of $65, $45, and $38 per million credits on Pro, Startup, and Scale.
- **Audio Realism Bench:** Design Arena's blind listening test, with rank among 13 AI voices and an Elo score. The scored model is named, and Google's row is Gemini TTS, not Chirp 3: HD.

</details>

### 1. Bland Speech

Bland Speech is a text to speech model that Bland runs itself, sold as a studio and an API. Audio streams over HTTP at POST /v2/tts or a WebSocket at /v2/tts/ws, where a new context immediately interrupts the turn that is playing, and output is mu-law at 8 kHz or PCM from 8 to 48 kHz. Bland publishes one latency figure, 199ms to first audio on a streamed request, and both endpoints are part of Bland's [text to speech API](/text-to-speech-api).

Delivery is steered inline on current-generation voices: [laughs], [sighs], and [say warmly] sit before the words they shape, and expressiveness and stability run from 0.0 to 1.0. 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 after Bland's consent and rights review. 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).

<div class="cmp-proscons cmp-proscons-own"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>Bland Speech v3 ranked first of 13 AI voices in Audio Realism Bench results of 4 August 2026</li><li>One published latency figure and one standard per-character rate</li><li>Mu-law at 8 kHz and PCM from 8 to 48 kHz for phone lines</li><li>An OpenAI-compatible endpoint, plus Pipecat and LiveKit (Python) plugins</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>

**Pricing:** $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. Enterprise pricing is custom, set with Bland's sales team. The [Bland Speech](/speech) page has a studio for trying a voice.

### 2. Cartesia

Cartesia sells Sonic 3.6, a model that speaks 44 languages from one model id, and says it delivers first audio in under 90ms. The model suits a team that wants a published latency claim on one model and will manage monthly credit allowances: one credit buys one character, overages are opt-in, and credits roll over up to twice the monthly allotment.

<div class="cmp-proscons" data-rival="cartesia"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>Cartesia says first audio arrives in under 90ms</li><li>SOC 2 Type II, with BAAs and DPAs on Enterprise</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>Credits, rollover limits, and overage settings to manage each month</li><li>Emotion controls are beta and English only</li></ul></div></div>

**Pricing:** Free is $0 a month with 20,000 credits, Pro $5 with 100,000, Startup $49 with 1.25 million, and Scale $299 with 8 million, with Enterprise custom.

### 3. ElevenLabs

ElevenLabs launched Eleven v4 and v4 Turbo on 28 September 2026, and the v4 family speaks 90+ languages. Eleven v4 Turbo is the real-time model, with a median latency of about 100ms excluding application and network time. The platform suits a team that wants a large voice library and will manage plan tiers, and ElevenLabs offers BAAs on Enterprise.

<div class="cmp-proscons" data-rival="elevenlabs"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>90+ languages across the Eleven v4 family</li><li>Audio tags such as [whispering] and [laughing] on Eleven v4</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>Eleven v4 bills $0.08 per 1,000 characters, twice Eleven v4 Turbo's $0.04</li><li>The standard text to speech WebSocket doesn't take Eleven v3 or v4</li></ul></div></div>

**Pricing:** $0.08 per 1,000 characters for Eleven v4 and $0.04 for Eleven v4 Turbo and Flash. Plans run Starter $6 a month, Creator $22, Pro $99, Scale $299, and Business $990, with Enterprise custom.

### 4. Murf

Murf's API has two models: Falcon 2.1 for streaming agents and Gen2 for studio-quality narration. Murf says Falcon 2.1 reaches first audio in under 100ms, and an older Murf page claims 55ms of model inference and 130ms end to end. The API suits a team that will accept a single conversational style to pay the lowest per-character rate in this list.

<div class="cmp-proscons" data-rival="murf"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>Falcon 2.1 at $0.01 per 1,000 characters</li><li>$10 of free API credit, and purchased credit never expires</li><li>Official LiveKit and Pipecat integrations</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>Falcon 2.1 voices speak in one style, Conversational</li><li>Instant voice cloning is on Enterprise only</li></ul></div></div>

**Pricing:** Falcon 2.1 costs $0.01 per 1,000 characters and Gen2 $0.03, with $10 of free API credit. Enterprise is custom.

### 5. Google Cloud

Google Cloud Text-to-Speech has Chirp 3: HD voices and Gemini TTS models, with 380+ voices across 75+ languages and variants, and Google describes its streaming as ultra-low-latency. The service suits a team already on Google Cloud, because Text-to-Speech is covered by Google's HIPAA BAA and sits in its SOC 2 scope. It returns mu-law and A-law.

<div class="cmp-proscons" data-rival="google"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>BAA coverage under Google Cloud's HIPAA agreement</li><li>Bidirectional streaming that takes text and returns audio</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>Chirp 3: Instant Custom Voice is closed to new access requests</li><li>Gemini voices bill per token and have no free usage</li></ul></div></div>

**Pricing:** Chirp 3: HD is $30 per million characters after a free first million each month. Gemini 2.5 Pro TTS is $1.00 per million input tokens and $20.00 per million audio tokens, at 25 audio tokens per second of speech.

## Which Inworld TTS alternative sounds most human?

Bland Speech v3 ranked first of 13 AI voices at 1365 in Design Arena's Audio Realism Bench results of 4 August 2026, and the human recording scored 1500. Native speakers hear two recordings of the same line and pick the one that sounds more like a real person, and Elo, the rating system chess uses, turns the picks into a score.

Inworld's TTS-1.5 Max ranked 13th at 763, 602 points behind Bland Speech v3. Google's Gemini 2.5 Pro TTS ranked 5th at 1109, Cartesia's Sonic 3.5 ranked 6th at 1096, and ElevenLabs' Eleven v3 ranked 12th at 938, while Murf wasn't in those results. Inworld's TTS-2 (5 May 2026) and TTS-2 Flash (9 August 2026) weren't scored, and Cartesia's Sonic 3.6 and ElevenLabs' Eleven v4 were released after those results. A score for an older Inworld model doesn't settle how TTS-2 sounds, so a team should listen to TTS-2 on its own script.

## Which Inworld TTS alternative should I pick?

A team that needs a voice for live calls should pick Bland Speech: Bland Speech v3 ranked first of 13 AI voices in those 4 August 2026 results, it returns mu-law at 8 kHz, and Bland publishes one latency figure and one standard rate.

<div class="cmp-usewhen"><div class="cmp-usewhen-own"><p><strong>Use Bland Speech when</strong></p><ul><li>A team will put a voice on phone calls and plan around one published latency figure</li><li>A team will bill by the character with no monthly plan</li></ul></div><div><p><strong>Use another option when</strong></p><ul><li>A team will manage monthly credit allowances for a published sub-90ms claim: Cartesia</li><li>A team will choose from a large voice library across 90+ languages and manage plan tiers: ElevenLabs</li><li>A team will accept a single conversational style to pay $10 per million characters: Murf</li><li>A team already runs on Google Cloud and wants one BAA: Google Cloud</li><li>A team will choose between TTS-2 Flash for speed and TTS-2 for steering, on a plan that lowers the rate: Inworld</li></ul></div></div>

Inworld's TTS-2 Flash costs $15 per million characters and publishes lower latency numbers than Bland Speech's 199ms, so a team that ranks those two over how a voice sounds will find Inworld cheaper and faster on paper.

## Frequently asked questions

### Which Inworld TTS latency figure should a team trust?

None of them as a promise. Inworld says TTS-2 Flash returns P90 time to first audio byte in 20ms server-side excluding network time, 45ms at the median on a Cekura benchmark, and as low as 50ms on Coval, so a team should measure end to end on its own network and script.

### How much does Inworld TTS cost?

On-Demand is $0 a month with TTS-2 at $25 per million characters and TTS-2 Flash at $15, and paid plans lower both, to $12.50 and $7 on the $1,500 Growth plan. Enterprise goes as low as $5 for TTS-2 and sub-$5 for Flash, with price match available.

### What is the best Inworld TTS alternative for phone calls?

Bland Speech is the best Inworld TTS alternative for phone calls, because it returns mu-law at 8 kHz, accepts text from a language model over a WebSocket, and ranked first of 13 AI voices in Design Arena's results of 4 August 2026.

### Does Inworld TTS return phone audio?

Yes. Inworld returns mu-law and A-law at 8 kHz, and has a guide for powering phone calls through Twilio ConversationRelay.

### How many languages does Inworld TTS cover?

Inworld says "500+ languages and dialects" in one place and "200+ languages and locales" in another, and Inworld's TTS-2 accepts 623 language and locale codes covering 470 languages in three tiers. ElevenLabs' v4 family speaks 90+ languages, Google Cloud covers 75+ languages and variants, and Cartesia's Sonic 3.6 speaks 44.

### Was Inworld's current model scored in the Audio Realism Bench results?

No. Design Arena's results of 4 August 2026 scored Inworld TTS-1.5 Max, which ranked 13th at 763, and TTS-2 had launched on 5 May 2026 without appearing in them.
