# 5 Gradium alternatives in 2026 for voice agents on live phone calls

By Samuel Oram-Jones

Gradium publishes two latency figures for its text to speech: a product headline of "under 50ms" and a developer introduction that says the expected time to first token is below 300ms when streaming, from servers in Europe and the US. Both are Gradium's own claims, and a buyer comparing providers on first audio is comparing claims that each company measures its own way, as of November 2026.

Bland publishes 199ms to first audio for Bland Speech, which isn't the lowest figure among these five, so the case for Bland Speech rests on something else. 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 Bland sells it at $0.04 per 1,000 characters with no monthly plan.

## Why do people look for Gradium alternatives?

People look for Gradium alternatives when a voice agent needs more than five languages or credits that don't expire at the end of each month, because Gradium sells five languages on monthly credit plans whose credits don't roll over.

Gradium sells real-time voice AI for voice agents: text to speech, speech to text, live translation, and cloning. It supports English, French, German, Spanish, and Portuguese, and Gradium reports its own hard-case pass rate of 81.0% on 500 sentences in five languages, against 75.1% for Cartesia Sonic 3.6 and 65.4% for ElevenLabs v3 on the same benchmark.

Gradium bills one credit per character. Free is $0 a month with 45,000 credits, XS is $12 with 225,000, S is $39 with 900,000, M is $312 with 9 million, and L is $1,480 with 45 million. Gradium prices a million credits at $58 on XS and $48 on S, falling to $38 on M and $36 on L, so Gradium costs more per character than Bland Speech's $40 per million at the small tiers and less at the large ones. Plan credits must be used within the month and don't roll over, commercial use starts on XS, and a Pro voice clone costs a one-time 1 million credits to train and then 1.2 credits per character. Gradium says teams under HIPAA usually take its VPC route.

## What should I check in a Gradium alternative?

A buyer should check what a first-audio figure measures, how many languages a call needs, whether credits expire, and which compliance paperwork the vendor signs.

A first-audio figure is the time from sending text to receiving the first sound, and each company here measures its own, so a team times a streamed request from its own servers with its own text. Languages come next: Cartesia says Sonic 3.6 speaks 44 languages, and ElevenLabs says its v4 family supports 90+. Billing shape matters too, because Gradium's plan credits expire each month while Bland Speech has no monthly plan.

Bland is SOC 2 Type II audited and HIPAA compliant with a signed BAA, a business associate agreement, the contract HIPAA requires before a vendor handles patient data. Gradium holds ISO 27001:2022, SOC 2 Type 2, and HDS v2, a French health-data hosting certification, and says teams under HIPAA usually take its VPC route.

## Which Gradium alternative sounds most human?

Bland Speech v3 ranked first of 13 AI voices in Design Arena's Audio Realism Bench results of 4 August 2026, at 1365, and Gradium wasn't in those results, so they place no Gradium voice beside it.

Cartesia's Sonic 3.5 ranked 6th at 1096, ElevenLabs' Eleven v3 ranked 12th at 938, and Inworld's one entry was TTS-1.5 Max, ranked 13th at 763, which isn't its current TTS-2 line. Cartesia's Sonic 3.6 came out on 27 August 2026 and ElevenLabs' Eleven v4 on 28 September 2026, so neither was scored, and Deepgram wasn't in those results. Each model had one female and one male American English voice, and listeners heard two recordings of the same text, blind, and picked the one that sounded more like a real person.

## What are the best Gradium alternatives?

The best Gradium alternatives are Bland Speech, Cartesia, Deepgram, Inworld, and ElevenLabs, with Bland Speech first because it pairs the realism result of 4 August 2026 with phone-ready audio and a per-character rate with no plan attached.

| Alternative | First audio, as each company publishes it | Published price | Audio Realism Bench, 4 August 2026 |
| --- | --- | --- | --- |
| Bland Speech | 199ms | $40 per million characters, pay as you go | Bland Speech v3: ranked 1st, 1365 |
| Cartesia | Under 90ms | 1 credit per character, on plans from $5 a month | Sonic 3.5: ranked 6th, 1096 |
| Deepgram | As low as 80ms | $15 to $45 per million characters by model | Not in those results |
| Inworld | As low as 50ms on TTS-2 Flash | $15 to $25 per million characters on-demand | TTS-1.5 Max: ranked 13th, 763 |
| ElevenLabs | About 150ms median on Eleven v4 Turbo | $40 to $80 per million characters by model, on plans from $6 a month | Eleven v3: ranked 12th, 938 |

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

- **First audio, as each company publishes it:** the time to the first sound of a streamed request. Each figure is the company's own claim, measured its own way, and none was measured by Bland.
- **Published price:** each company's own list rate. ElevenLabs charges $0.08 per 1,000 characters on v4 and $0.04 on v4 Turbo and Flash. Deepgram charges $0.015, $0.030, and $0.045 per 1,000 on Aura-1, Aura-2, and Flux TTS. Inworld's On-Demand rates are $25 per million for TTS-2 and $15 for TTS-2 Flash.
- **Audio Realism Bench, 4 August 2026:** the model Design Arena scored, its place among the 13 AI voices, and its Elo score, a rating built from blind head-to-head votes. Inworld's entry is an older model than the one it sells now.

</details>

### 1. Bland Speech

Bland Speech is Bland's own text to speech model, v3, sold as a studio plus one API and built for phone calls. A WebSocket takes text as a language model writes it, a new context interrupts the turn that is playing, and audio returns as mu-law at 8 kHz or PCM from 8 to 48 kHz. Pipecat has a first-party Bland service, and LiveKit Agents has a Python plugin. More in the [text to speech API](/text-to-speech-api) overview.

An instant clone takes one clean sample of about 10 seconds, a professional clone is fine-tuned on 30 minutes or more of audio after Bland's consent and rights review, and every clone is private to the customer's organization. Bland is also FedRAMP 20x Class C certified at Moderate, granted on 17 November 2026, and Bland's trust page says speech synthesis runs inside that boundary. Bland Speech suits a team running live phone calls that wants a realistic voice at $0.04 per 1,000 characters with no credits to expire.

<div class="cmp-proscons cmp-proscons-own"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>Ranked first of 13 AI voices on 4 August 2026, at 1365</li><li>Pay as you go with no monthly credits to expire, and 50,000 free characters once</li><li>SOC 2 Type II, HIPAA compliant with a signed BAA, and FedRAMP 20x Class C at Moderate for Bland</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>More than a prototype needs, because Bland Speech is built for production calls</li></ul></div></div>

### 2. Cartesia

Cartesia sells Sonic 3.6, which Cartesia says speaks 44 languages and delivers first audio in under 90ms. Text goes in over a WebSocket, audio returns as mu-law at 8 kHz, and plans run on credits: Pro is $5 a month for 100,000, Startup is $49 for 1.25 million, and Scale is $299 for 8 million. Cartesia's credits roll over up to twice the monthly allotment. Cartesia suits a team that wants a wide language range at the latency Cartesia publishes and will manage a credit allowance.

<div class="cmp-proscons" data-rival="cartesia"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>44 languages on Sonic 3.6</li><li>Credits roll over up to 2x the monthly allotment</li><li>SOC 2 Type II, and BAAs on Enterprise</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>Overages are opt-in, and requests stop at zero credits with them off</li><li>Emotion controls are in beta and English only</li></ul></div></div>

### 3. Deepgram

Deepgram sells Flux TTS, Aura-2, and Aura-1 at $0.045, $0.030, and $0.015 per 1,000 characters on pay as you go, with a $200 credit that needs no card. Deepgram says Flux streams in as low as 80ms at 24 kHz, serves it over a WebSocket, and returns mu-law at 8 kHz. Deepgram suits a team that already uses Deepgram for speech to text and wants both from one vendor, with self-hosted deployment available.

<div class="cmp-proscons" data-rival="deepgram"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>A $200 credit with no card and no expiry</li><li>A delivery-register dial from calm to animated on Flux</li><li>SOC 2 Type 2 and a BAA for Enterprise customers</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>SSML is stripped, not supported</li><li>Flux TTS costs three times Aura-1 per character</li></ul></div></div>

### 4. Inworld

Inworld sells Realtime TTS-2 at $25 per million characters and TTS-2 Flash at $15 on its $0-a-month On-Demand plan, aimed at consumer-facing apps. Inworld says TTS-2 Flash reaches first audio in as little as 50ms, instant clones need as little as 3 seconds, and both models render 50 non-verbal sounds such as [laugh] and [sigh]. Text can stream in over a WebSocket as a language model writes it. Inworld suits a team that wants a low per-character rate and consumer-style voices.

<div class="cmp-proscons" data-rival="inworld"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>$15 per million characters on TTS-2 Flash</li><li>SOC 2 Type II, with HIPAA and a BAA on Enterprise or as a Growth add-on</li><li>Mu-law at 8 kHz</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>Steering in plain words works on TTS-2 only</li><li>Zero data retention leaves out voice cloning samples</li></ul></div></div>

### 5. ElevenLabs

ElevenLabs sells Eleven v4 and Eleven v4 Turbo, launched on 28 September 2026, with API use billed in US dollars per character at $0.08 per 1,000 on v4 and $0.04 on v4 Turbo and Flash. ElevenLabs says Eleven v4 Turbo's median time to first speech is about 150ms, and its v4 family covers 90+ languages. Audio tags such as [whispering] steer delivery. ElevenLabs suits a team that wants a broad language range and will manage a plan, with instant cloning from Starter and professional cloning from Creator.

<div class="cmp-proscons" data-rival="elevenlabs"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>90+ languages on the v4 family</li><li>A voice-captcha check for professional clones</li><li>SOC 2 Type 2, ISO 27001:2022, and BAAs for qualifying healthcare enterprises</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>The standard text to speech WebSocket doesn't take Eleven v3 or v4</li><li>Free users are limited to non-commercial use</li></ul></div></div>

## Which Gradium alternative should I pick?

First-audio claims from 49ms to 199ms can't settle this, because every figure is a company's own, so the better question is which voice a caller hears and which bill a team accepts.

Gradium's M and L tiers cost less per character than Bland Speech, at $38 and $36 per million against $40, so a team at that volume with callers only in Gradium's five languages pays less by staying. What moving buys is the voice that ranked 1st at 1365 on 4 August 2026 and Bland credits that never expire.

Bland Speech and Gradium go head to head in [Bland Speech vs Gradium](/compare/bland-speech-vs-gradium) and the [Gradium review](/blog/gradium-review), and the wider field is in the [text to speech comparison hub](/compare/text-to-speech) and the [alternatives family](/compare/text-to-speech/alternatives). Bland Speech itself is at [bland.ai/speech](/speech).

## Frequently asked questions

### What is the best Gradium alternative?

Bland Speech is the best Gradium alternative for live phone calls, because Bland Speech v3 ranked first of 13 AI voices in Design Arena's results of 4 August 2026 at 1365 and bills $0.04 per 1,000 characters with no monthly plan.

### How does Gradium price text to speech?

Gradium charges one credit per character on monthly plans: Free is $0 for 45,000 credits, XS is $12, S is $39, M is $312, and L is $1,480. Plan credits don't roll over, and Gradium prices a million credits from $58 on XS down to $36 on L.

### Which Gradium alternative covers more than five languages?

Cartesia says Sonic 3.6 speaks 44 languages, and ElevenLabs says its v4 family supports 90+. Gradium supports English, French, German, Spanish, and Portuguese.

### Which Gradium alternative signs a BAA for HIPAA?

Bland is HIPAA compliant with a signed BAA. ElevenLabs offers BAAs to qualifying healthcare enterprises, Cartesia and Deepgram sign them on Enterprise, and Inworld offers one on Enterprise or as a Growth add-on. Gradium says teams under HIPAA usually take its VPC route.

### Which Gradium alternative has the lowest first-audio figure?

Each company publishes its own claim, and none is a race Bland ran: Inworld says as low as 50ms for TTS-2 Flash, Deepgram as low as 80ms, Cartesia under 90ms, ElevenLabs about 150ms for Eleven v4 Turbo, and Bland 199ms. Gradium publishes 49ms at the 50th percentile on a Speko benchmark.

### Can a Gradium alternative return phone audio?

Yes, Bland Speech, Cartesia, Deepgram, Inworld, and ElevenLabs each return mu-law at 8 kHz, as Gradium does with its `ulaw_8000` format. Bland Speech also returns PCM from 8 to 48 kHz.
