# OpenAI TTS vs Google text to speech (2026): Which endpoint is safe to build on?

By Samuel Oram-Jones

Neither OpenAI nor Google keeps its newest voices on a plain text to speech endpoint any more. As of November 2026, OpenAI says `tts-1`, `tts-1-hd`, and `gpt-4o-mini-tts` shut down on 6 January 2027 and points developers to `gpt-realtime-2.1-mini` on its Realtime API, which it says "isn't a supported model for `/v1/audio/speech`". Google says Gemini 3.8 Flash TTS and Flash-Lite TTS aren't available through the Cloud Text-to-Speech API, and Google has closed new access requests for Chirp 3: Instant Custom Voice.

A team picking between the two is choosing which migration to take on, and when. Bland Speech serves an OpenAI-compatible `/audio/speech` endpoint at a standard rate of $0.04 per 1,000 characters ($40 per million), pay as you go with no monthly plan, and Bland Speech v3 ranked first of 13 AI voices in Design Arena's Audio Realism Bench results of 4 August 2026.

## Which is better, OpenAI TTS or Google text to speech?

Google text to speech is the safer base for a new build, because every model on OpenAI's speech endpoint carries a shutdown date of 6 January 2027. OpenAI's endpoint is the simpler API for a prototype that only has to run until then, built by a team that already holds an OpenAI key.

For a phone agent, the question of sound and format settles it differently, and Bland Speech fits that job more directly than either.

| | Bland Speech | OpenAI TTS | Google text to speech |
|---|---|---|---|
| Status | Bland Speech v3 is current | `tts-1`, `tts-1-hd`, and `gpt-4o-mini-tts` shut down 6 January 2027 | Chirp 3: HD and Gemini-TTS in the Cloud API; Instant Custom Voice closed to new access |
| Price per million characters | $40, one standard rate | `tts-1` $15, `tts-1-hd` $30; `gpt-4o-mini-tts` by token | Chirp 3: HD $30, Neural2 $16, WaveNet and Standard $4, Studio $160; Gemini-TTS by token |
| Free allowance | 50,000 characters once | Pay per use | First 1 million Chirp 3: HD characters each month; new customers get up to $300 in credits |
| Phone audio | mu-law at 8 kHz, PCM 8 to 48 kHz | mp3, opus, aac, flac, wav, and 24 kHz pcm | MULAW and ALAW encodings |
| Streaming | HTTP chunks and WebSocket | HTTP chunked transfer; WebSocket only on the Realtime API | Bidirectional streaming in the client library |
| Steering delivery | Tags such as [laughs] and [say warmly] | A plain-language `instructions` field on `gpt-4o-mini-tts` | Bracket tags and a style prompt on Gemini-TTS; speaking rate and pause tags on Chirp 3: HD |
| Cloning | Instant from one clean sample of about 10 seconds | Custom voices for eligible customers, from samples of 60 seconds or less | Voice replication in Gemini 3.8 TTS, in Preview |
| Realism on 4 August 2026 | 1st of 13 at 1365 | GPT-4o Mini TTS 9th at 1011; GPT Realtime 2 4th at 1119 | Gemini 2.5 Pro TTS 5th at 1109; Gemini 3.1 Flash TTS 8th at 1037; Gemini 2.5 Flash TTS 10th at 985 |

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

- **Price per million characters:** Each company's rate converted to one unit. OpenAI and Google price their newer speech models by token instead, so a per-character figure doesn't exist for them.
- **Phone audio:** Mu-law and A-law are compressed 8-bit formats used on phone lines. A voice that outputs them directly needs no conversion before a call carries it.
- **Realism on 4 August 2026:** Position and Elo score in Design Arena's Audio Realism Bench results of that date. Elo is a rating system built from head-to-head votes, here from listeners choosing which clip sounds more like a person. Chirp 3: HD and the Gemini 3.8 models were not scored.

</details>

## Where do OpenAI TTS and Google text to speech fall short for a voice agent?

OpenAI TTS falls short on lifespan and phone formats, and Google text to speech falls short on spread, since its best voices sit across Chirp 3: HD, Gemini-TTS in the Cloud API, and Gemini 3.8 on a different API.

### OpenAI TTS

OpenAI announced the deprecation on 1 October 2026, and the shutdown on 6 January 2027 covers `tts-1`, `tts-1-hd`, `gpt-4o-mini-tts`, and its dated snapshots. The replacement runs on the Realtime API over WebRTC or WebSocket, which OpenAI describes as "a change of API workflow", so code that posts text to `/v1/audio/speech` and plays back the bytes has to be rewritten, not renamed.

The speech endpoint does not return mu-law, the format a phone line carries: its formats are mp3, opus, aac, flac, wav, and pcm at 24 kHz. The Realtime session accepts `audio/pcmu`, which is mu-law, as an output. OpenAI says `tts-1` gives lower latency than `tts-1-hd` at lower quality. Server-sent events are not supported on those two models. OpenAI says voices are optimized for English, and its usage policies require telling end users the voice is AI-generated.

Custom voices from recordings are limited to eligible customers. Each needs a consent recording with an approved phrase, a sample of 60 seconds or less, and counts toward 20 voices per organization.

<div class="cmp-proscons" data-rival="openai"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>tts-1 at $15 per million characters and tts-1-hd at $30</li><li>A plain-language instructions field on gpt-4o-mini-tts for accent, tone, and whispering</li><li>SOC 2 Type 2 for its API and a Business Associate Agreement offered</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>Every speech endpoint model shuts down on 6 January 2027</li><li>The replacement is a different API, not a drop-in</li><li>No mu-law on the speech endpoint</li><li>Custom voices only for eligible customers</li></ul></div></div>

OpenAI TTS is a poor fit for:

- A new build that has to run past 6 January 2027 without a rewrite
- A phone line that wants mu-law straight from the speech endpoint
- A team that wants cloning without a sales conversation

### Google text to speech

Google's spread is the catch. Chirp 3: HD is $30 per million characters, with the first million free each month, and takes no emotion control, only `speaking_rate` from 0.25 to 2.0 and pause tags. Chirp 3: HD streaming does not support SSML. The Gemini-TTS models respond to bracket tags such as [sigh], [whispering], and [short pause] and to a style prompt, and they bill by token, with 25 audio tokens per second of audio. Google warns that vocalized-adjective tags such as [scared] are spoken aloud, which it calls "likely an undesired side effect for most use cases".

Prices are moving too. Gemini 3.8 Flash TTS output costs $9.00 per million audio tokens through 31 December 2026 and $18.00 from 1 January 2027, and no Gemini-TTS model has free usage.

Cloning is mid-transition. Chirp 3: Instant Custom Voice, priced at $60 per million characters, no longer takes new access requests, and Google points to voice replication in Gemini 3.8 TTS, which is in Preview.

<div class="cmp-proscons" data-rival="google"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>380+ voices across 75+ languages and variants</li><li>MULAW and ALAW encodings, and bidirectional streaming</li><li>First million Chirp 3: HD characters free each month</li><li>Text to Speech is covered by Google Cloud's HIPAA Business Associate Agreement</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>Best voices spread across Chirp 3: HD, Gemini-TTS, and the Agent Platform API</li><li>Gemini-TTS has no free usage and bills by token</li><li>Instant Custom Voice is closed to new access</li><li>No emotion control on Chirp 3: HD</li></ul></div></div>

Google text to speech is a poor fit for:

- A team that wants one price list and one API for its best voices
- A cloned-voice project that starts today on Instant Custom Voice
- A budget that cannot absorb Gemini 3.8 Flash TTS output prices doubling on 1 January 2027

Each rival also has its own head-to-head, [Bland Speech vs OpenAI TTS](/compare/bland-speech-vs-openai-tts) and [Bland Speech vs Google text to speech](/compare/bland-speech-vs-google-tts), and [OpenAI TTS alternatives](/blog/openai-tts-alternatives) names where OpenAI's speech customers can move.

## What do OpenAI TTS, Google text to speech, and Bland Speech cost?

At published per-character rates, OpenAI's `tts-1` is $15 per million, Google's Chirp 3: HD is $30, and Bland Speech is $40, with the token-priced models each working out to cents per minute.

Ten million characters a month costs $150 on `tts-1`, $300 on `tts-1-hd`, $270 on Chirp 3: HD after the first million free, $144 on Neural2 after its free million, and $400 on Bland Speech. Token pricing reads differently: Google's Gemini 2.5 Flash TTS output of $10.00 per million audio tokens works out to $0.015 for each minute of audio before input tokens, and its Pro and 3.1 Flash models at $20.00 work out to $0.03. Bland's own worked example puts a 10-minute narration at $0.36.

Bland Speech costs more per character than either rival's standard voices. Phone minutes, taxes, and the language model that writes the text come on top at all three.

## Which sounds more human, OpenAI TTS, Google text to speech, or Bland Speech?

In Design Arena's Audio Realism Bench results of 4 August 2026, Bland Speech v3 ranked first of 13 AI voices at 1365, against a human recording at 1500, while OpenAI's GPT Realtime 2 ranked 4th at 1119 and GPT-4o Mini TTS 9th at 1011. Google's Gemini 2.5 Pro TTS ranked 5th at 1109, Gemini 3.1 Flash TTS 8th at 1037, and Gemini 2.5 Flash TTS 10th at 985.

Chirp 3: HD, the Gemini 3.8 models, and OpenAI's `gpt-realtime-2.1-mini` weren't in those results. Design Arena has vetted native speakers hear two recordings of one transcript, blind and in random order, and pick the one that sounds more like a real person; the full method is in [Design Arena's article](https://notes.designarena.ai/audio-realism-benchmark-measuring-realism-in-audio-models/).

The scored OpenAI model most comparable to the speech endpoint, GPT-4o Mini TTS, is the one on the shutdown list. Google's best scored model is a Gemini model that bills by token, not by character.

## What does Bland Speech offer a team leaving OpenAI TTS or weighing Google text to speech?

Bland Speech offers a team leaving OpenAI TTS or weighing Google text to speech the OpenAI request shape plus what a phone agent needs: mu-law at 8 kHz, a standard per-character rate, and Bland Speech v3's first place in the 4 August 2026 realism results.

Bland's `/audio/speech` endpoint implements OpenAI's contract, so a team switching clients changes the base URL, the API key, and the model. Four things differ from OpenAI: `speed` accepts only 1.0, `instructions` is accepted and ignored, OpenAI model IDs such as `tts-1` and `gpt-4o-mini-tts` are not aliased, and `sse` returns a 400 error. Bland's native endpoints are `POST /v2/tts` and a WebSocket at `/v2/tts/ws` that takes text as a language model writes it and lets a new context interrupt the turn that is playing. The [text to speech API](/text-to-speech-api) page has the detail, and [how to migrate from OpenAI TTS](/blog/migrate-from-openai-tts) walks the move.

Bland publishes 199ms to first audio. Bland Speech has a first-party Pipecat service and a LiveKit Agents plugin for Python. An instant clone needs one clean sample of about 10 seconds, a professional clone is fine-tuned on 30 minutes or more of audio and goes through Bland's consent and rights review, and every clone is private to the customer's organization. Bland is SOC 2 Type II audited and HIPAA compliant with a signed BAA.

<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 in Design Arena's results of 4 August 2026</li><li>An OpenAI-compatible /audio/speech endpoint, plus native HTTP and WebSocket endpoints</li><li>8 kHz mu-law output and PCM from 8 to 48 kHz</li><li>One standard rate with no monthly plan</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>

More matchups sit in the [text to speech hub](/compare/text-to-speech) and the [rival vs rival family](/compare/text-to-speech/rival-vs-rival), and Bland Speech itself is at [bland.ai/speech](/speech).

## Frequently asked questions

### Is OpenAI TTS better than Google text to speech?

For a build that has to last, Google text to speech is better, because OpenAI's `tts-1`, `tts-1-hd`, and `gpt-4o-mini-tts` shut down on 6 January 2027. For a quick prototype on an OpenAI key, OpenAI's endpoint is simpler until then.

### Is OpenAI's text to speech API shutting down?

Yes, the three speech endpoint models are. OpenAI deprecated `tts-1`, `tts-1-hd`, and `gpt-4o-mini-tts` on 1 October 2026 and set the shutdown for 6 January 2027. The replacement, `gpt-realtime-2.1-mini`, runs on the Realtime API, which OpenAI says is a change of API workflow.

### Which is cheaper, OpenAI TTS or Google text to speech?

At per-character rates, OpenAI's `tts-1` at $15 per million undercuts Google's Chirp 3: HD at $30, though Google gives the first million characters free each month. Google's WaveNet and Standard voices cost $4 per million. Bland Speech costs $40 per million.

### Can OpenAI TTS or Google clone a voice?

OpenAI offers custom voices to eligible customers, from recordings of 60 seconds or less with a consent recording. Google has closed new access requests for Chirp 3: Instant Custom Voice and points to voice replication in Gemini 3.8 TTS, which is in Preview. Bland Speech clones from one clean sample of about 10 seconds.

### Do OpenAI TTS and Google output mu-law for phone calls?

Google does, and OpenAI's speech endpoint does not. Google offers MULAW and ALAW encodings, and OpenAI's speech endpoint returns mp3, opus, aac, flac, wav, or 24 kHz pcm. OpenAI's Realtime API accepts `audio/pcmu` as an output. Bland Speech outputs 8 kHz mu-law and PCM from 8 to 48 kHz.

### Do OpenAI and Google support HIPAA?

Both support it through a Business Associate Agreement. OpenAI offers a BAA to customers, and Google Cloud enters into BAAs, with Text to Speech in its covered products list. Bland is HIPAA compliant with a signed BAA.

### Can Bland Speech replace OpenAI TTS with a code change?

For most `/audio/speech` calls, yes: the base URL, API key, and model change. Bland's endpoint accepts `speed` of 1.0 only, ignores `instructions`, and returns 400 for `sse`, so a client that relies on those needs adjusting.
