# 6 OpenAI TTS alternatives in 2026 before the 6 January 2027 shutdown

By Samuel Oram-Jones

OpenAI's text to speech endpoint stops serving tts-1, tts-1-hd, and gpt-4o-mini-tts on 6 January 2027, and the model OpenAI names as their replacement can't be called from that endpoint. OpenAI's own words are that moving to gpt-realtime-2.1-mini "is a change of API workflow", so a team that sends text to /v1/audio/speech and saves an audio file rewrites that code either way, as of November 2026.

Six options follow: Bland Speech, ElevenLabs, Google Cloud, Cartesia, Deepgram, and Amazon Polly. 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 a request shaped like OpenAI's can reach it with three fields changed. 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 OpenAI TTS](/compare/bland-speech-vs-openai-tts) sets the two side by side.

## Why do teams look for OpenAI TTS alternatives?

Teams look for OpenAI TTS alternatives because OpenAI announced on 1 October 2026 that its three speech models shut down on 6 January 2027, and the replacement isn't supported on the endpoint they share.

That endpoint takes text and returns mp3, opus, aac, flac, wav, or raw PCM, and the PCM is 24 kHz, so none of its formats is the 8 kHz mu-law (a compressed 8-bit format that phone networks carry) a phone line wants. tts-1 costs $15 per million characters and tts-1-hd $30. gpt-4o-mini-tts bills tokens instead, at $0.60 per million text input tokens and $12 per million audio output tokens.

The replacement, gpt-realtime-2.1-mini, runs on the Realtime API over WebRTC or WebSocket, with audio priced at $10 per million input tokens and $20 per million output tokens. Its session config accepts audio/pcmu output, the telephony name for mu-law, but the work is a new integration built on sessions.

Delivery is a second cost. gpt-4o-mini-tts took a plain-language instructions field for tone, accent, and whispering, and that field belongs to a model that shuts down on the same day as the other two, so a team that tuned its voice with instructions retunes it elsewhere. OpenAI's strengths are a SOC 2 Type 2 examination of its API and a BAA (business associate agreement, the contract HIPAA requires).

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

A buyer should check how much code changes, how text streams in, what audio comes out, and what each million characters costs, because those four answers separate a three-field edit from a new integration.

Code change comes first. Bland Speech's POST /v2/audio/speech follows OpenAI's audio speech contract, and Bland says to "change the base URL, API key, and model". The endpoint has catches: speed accepts only 1.0, the instructions field is accepted and ignored, and a request that still says tts-1 returns a model_not_found error. ElevenLabs, Cartesia, Deepgram, Google Cloud, and Amazon Polly each use their own request shapes, such as Cartesia's /tts/bytes and Deepgram's /v2/speak, so a move to one of them means rewriting calls.

Streaming decides how a reply sounds. A WebSocket (a connection that stays open so text goes in and audio comes out as both are produced) lets a voice agent send a language model's words as they arrive. Bland Speech's WebSocket also lets a new context interrupt the turn that is playing, which is how a caller talking over the agent gets handled.

Phone audio narrows the list little, because all six alternatives offer a mu-law format. Price counting varies more, from Cartesia's monthly credits to Google Cloud's per-token Gemini voices.

## What are the best OpenAI TTS alternatives?

The best OpenAI TTS alternatives are Bland Speech, ElevenLabs, Google Cloud, Cartesia, Deepgram, and Amazon Polly, with Bland Speech first for phone calls and voice agents because of its realism result, an endpoint that follows OpenAI's contract, and one standard per-character rate.

| Alternative | Best for | Price per million characters | Audio Realism Bench, 4 August 2026 |
| --- | --- | --- | --- |
| Bland Speech | Phone calls and an OpenAI client kept as is | $40, pay as you go | Bland Speech v3: ranked 1st, 1365 |
| ElevenLabs | Audio tags and 90+ languages | $40 to $80, by model | Eleven v3: ranked 12th, 938 |
| Google Cloud | Teams already on Google Cloud | $30 for Chirp 3: HD, first million free each month | Gemini 2.5 Pro TTS: ranked 5th, 1109 |
| Cartesia | First audio Cartesia puts under 90ms | 1 credit per character, $38 to $65 per million credits past a plan | Sonic 3.5: ranked 6th, 1096 |
| Deepgram | Pay by usage with a $200 free credit | $45 for Flux TTS, $30 for Aura-2 | Wasn't in those results |
| Amazon Polly | Teams already on AWS | $4 for Standard, $16 for Neural, $30 for Generative | Wasn't in those results |

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

- **Price per million characters:** each company's published rate with the model or plan named.
- **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 Cloud'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, with a studio to try voices in and an API to build on. Three routes reach it: POST /v2/tts for complete text, WSS /v2/tts/ws for streamed text, and POST /v2/audio/speech, which follows OpenAI's contract and maps OpenAI's voice names onto Bland voices: alloy, ash, coral, and fable speak as River; amber, august, lily, nova, and shimmer as Karen; and ballad, blue, echo, onyx, sage, and verse as Matthew. Bland publishes 199ms to first audio, and all three routes belong to Bland's [text to speech API](/text-to-speech-api).

Performance tags in square brackets, such as [laughs] or [say warmly], take over the job of OpenAI's instructions field, so a prompt that said "speak warmly" becomes a tag in the script.

<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>An endpoint that follows OpenAI's contract, with OpenAI voice names mapped to Bland voices</li><li>Mu-law at 8 kHz and PCM from 8 to 48 kHz for phone lines</li><li>One standard per-character rate with no monthly plan to renew</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 the studio, and [how to migrate from OpenAI TTS](/blog/migrate-from-openai-tts) has the steps.

### 2. ElevenLabs

ElevenLabs launched Eleven v4 and Eleven v4 Turbo on 28 September 2026, and the v4 family speaks 90+ languages. The API suits a team that wants audio tags: [whispering], [shouting], and [laughing] work on Eleven v4, which is the nearest match to what OpenAI's instructions field did. A move from OpenAI means rewriting calls, because ElevenLabs' routes are its own.

<div class="cmp-proscons" data-rival="elevenlabs"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>Audio tags such as [whispering] and [shouting] on Eleven v4</li><li>90+ languages across the Eleven v4 family</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 WebSocket doesn't take Eleven v3 or v4</li><li>The Free plan is non-commercial, and a commercial licence starts on Starter</li></ul></div></div>

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

### 3. Google Cloud

Google Cloud Text-to-Speech sells Chirp 3: HD voices alongside Gemini TTS models priced per token, across 75+ languages and variants. The service suits a team already on Google Cloud, because Text-to-Speech is among the products Google's HIPAA BAA covers. Gemini TTS takes bracket tags such as [sigh] and [short pause], and Google describes a style prompt as "the primary driver of the overall emotional tone and delivery".

<div class="cmp-proscons" data-rival="google"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>Text-to-Speech sits inside Google Cloud's HIPAA BAA and SOC 2 scope</li><li>The first million Chirp 3: HD characters are free each month</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>Gemini voices bill per token, so a per-character budget needs a conversion</li><li>Gemini 3.8 Flash TTS reaches a team through the Agent Platform API, not the Cloud Text-to-Speech API</li><li>Chirp 3: HD streaming doesn't support SSML tags</li></ul></div></div>

**Pricing:** Chirp 3: HD costs $30 per million characters after a free first million each month. Gemini 2.5 Flash TTS costs $0.50 per million text input tokens and $10.00 per million audio output tokens, and Google counts 25 audio tokens per second of speech, so output alone runs about $0.015 per minute of audio.

### 4. Cartesia

Cartesia sells Sonic 3.6, a model that speaks 44 languages, with a current snapshot dated 27 August 2026. Cartesia says Sonic delivers first audio in under 90ms. The platform suits a team that will manage a monthly credit balance, because Cartesia bills one credit per character and, with overages off, stops requests when credits run out.

<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>Mu-law and A-law output at 8 kHz</li><li>One model across 44 languages</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>Instant voice cloning needs the Pro plan, and professional cloning needs Startup</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. Past an allowance, overages cost $65, $45, and $38 per million credits on Pro, Startup, and Scale.

### 5. Deepgram

Deepgram's Flux TTS is a voice agent model served on /v2/speak over WebSocket and batch REST, and Deepgram says responses start in as low as 80ms at 24 kHz under production load. The API suits a team that wants to pay by usage: new accounts get a $200 credit with no card and no expiry.

<div class="cmp-proscons" data-rival="deepgram"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>Deepgram says Flux TTS starts speaking in as low as 80ms</li><li>Mu-law and A-law output at 8 kHz</li><li>A $200 free credit with no card and no expiry</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>Flux TTS strips SSML markup with a warning and keeps synthesizing</li><li>Flux TTS costs $0.045 per 1,000 characters, 50% more than Aura-2's $0.030</li></ul></div></div>

**Pricing:** Flux TTS is $0.0450 per 1,000 characters on Pay as you go and Growth, and Aura-2 is $0.030 on Pay as you go and $0.027 on Growth. Growth starts at $4,000 a year prepaid, and Enterprise is custom.

### 6. Amazon Polly

Amazon Polly sells four voice engines, Standard, Neural, Long-Form, and Generative, with no plan tiers. The service suits a team already on AWS that needs plain voices at the lowest rate in this list, since Polly is a HIPAA Eligible Service under the AWS BAA.

<div class="cmp-proscons" data-rival="amazon-polly"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>Standard voices at $4 per million characters, with a 5 million character free tier each month</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>The Generative engine doesn't support speech marks or the Newscaster style</li><li>A custom Brand Voice goes through an AWS account manager</li></ul></div></div>

**Pricing:** Standard is $4.00 per million characters, Neural $16.00, Generative $30, and Long-Form $100.00. The free tier is 5 million Standard characters a month.

## Which OpenAI TTS alternative sounds most human on a call?

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. Elo is the rating system chess uses, applied to blind votes: a native speaker hears two recordings of one line and picks the more human.

OpenAI's own voices were in those results, with GPT Realtime 2 ranked 4th at 1119 and GPT-4o Mini TTS ranked 9th at 1011. Of the alternatives, 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 Deepgram, Amazon Polly, and gpt-realtime-2.1-mini weren't in them. Cartesia's Sonic 3.6 (27 August 2026) and ElevenLabs' Eleven v4 (28 September 2026) came after them and weren't scored.

A blind score says how a voice sounds on a clean recording, so a team should still play its own script through 8 kHz mu-law first.

## Which OpenAI TTS alternative should I pick?

The shutdown makes the code change the cost that decides this, and for a phone or voice agent the smallest change is to Bland Speech: the OpenAI-shaped route keeps an existing client, mu-law at 8 kHz covers the phone line, and Bland Speech v3 ranked first of 13 AI voices in those 4 August 2026 results.

<div class="cmp-usewhen"><div class="cmp-usewhen-own"><p><strong>Use Bland Speech when</strong></p><ul><li>A team will keep an OpenAI client and change its base URL, key, and model</li><li>A team will put a voice on phone calls and bill one standard per-character rate with no monthly plan</li></ul></div><div><p><strong>Use another option when</strong></p><ul><li>A team will tag delivery with [whispering] and manage plan tiers: ElevenLabs</li><li>A team already runs on Google Cloud and wants its BAA to cover speech: Google Cloud</li><li>A team will manage monthly credit allowances: Cartesia</li><li>A team will pay by usage from a $200 credit: Deepgram</li><li>A team already runs on AWS and needs $4 per million characters: Amazon Polly</li></ul></div></div>

On price the choice points the other way: tts-1 costs $15 per million characters and Bland Speech costs $40, so 10 million characters a month goes from $150 to $400, which a team should weigh against what a caller hears.

## Frequently asked questions

### What happens to OpenAI TTS on 6 January 2027?

OpenAI shuts down tts-1, tts-1-hd, and gpt-4o-mini-tts on 6 January 2027, after announcing it on 1 October 2026, and says its replacement isn't a supported model for /v1/audio/speech.

### What replaces OpenAI TTS?

OpenAI names gpt-realtime-2.1-mini on the Realtime API, which bills audio tokens at $10 per million in and $20 per million out. A team that wants to keep sending text and getting audio back can move to Bland Speech, ElevenLabs, Google Cloud, Cartesia, Deepgram, or Amazon Polly.

### Can Bland Speech replace OpenAI's speech endpoint with a base URL change?

Yes, with three catches. Bland Speech's POST /v2/audio/speech follows OpenAI's contract, so the base URL, API key, and model change, and the model becomes btts-3. Speed accepts only 1.0, the instructions field is ignored, and OpenAI's model ids such as tts-1 return an error, so delivery moves into bracket tags in the script.

### Which OpenAI TTS alternative costs the least?

Amazon Polly costs the least per character, with Standard voices at $4 per million characters against $15 for tts-1. Google Cloud's Chirp 3: HD is $30 per million after a free first million each month, and Bland Speech is $40 per million.

### Which OpenAI TTS alternatives return phone audio?

All six do: ElevenLabs offers ulaw_8000, Cartesia pcm_mulaw, Deepgram mulaw, Google Cloud MULAW, Amazon Polly mu-law, and Bland Speech mu-law at 8 kHz.

### Can an OpenAI custom voice move to another provider?

No, a cloned voice stays with the vendor that built it, so a team re-clones from the original recordings with consent. Bland Speech needs one clean sample of about 10 seconds for an instant clone and the customer's confirmation that the voice is theirs to use.
