OpenAI will shut down tts-1, tts-1-hd, and gpt-4o-mini-tts on 6 January 2027, and the model it names as their replacement, gpt-realtime-2.1-mini, isn't accepted by the /v1/audio/speech endpoint that all three share. OpenAI calls the move a change of API workflow, because the replacement runs over its Realtime API instead.
A team with voice agents or phone flows built on that endpoint has to move either way. Bland Speech implements the same audio speech request shape, so for some teams a move is a change of base URL, API key, and model name, with a few differences that matter, such as speed fixed at 1.0.
Which is better, Bland Speech or OpenAI TTS?
For phone calls and voice agents, Bland Speech is the better choice as of November 2026. Bland Speech v3 ranked first of 13 AI voices in Design Arena's Audio Realism Bench results of 4 August 2026, at 1365, while OpenAI's GPT-4o Mini TTS ranked 9th at 1011, and Bland Speech is a current model while OpenAI's three speech endpoint models are on a shutdown date.
OpenAI makes more sense for a team that is building a whole conversation on OpenAI's Realtime API rather than calling a speech endpoint, or one that wants the lower per-character rate of tts-1 for the months it has left. A team weighing a move can read the migration guide, or the wider OpenAI TTS alternatives list.
Use Bland Speech when
- An integration that calls /v1/audio/speech has to keep working after 6 January 2027 with the least code change
- Callers on a phone line should hear a voice that sounds like a person
- Audio goes onto a phone line as mu-law at 8 kHz, with no conversion step
- A script needs its delivery written inline, such as [laughs] or [say warmly]
- A team wants an instant voice clone from about 10 seconds of audio on its own account
Use OpenAI TTS when
- A team will run the whole call on OpenAI's Realtime API, over WebSocket or WebRTC, with the replacement model gpt-realtime-2.1-mini
- A team wants $15 per million characters on tts-1 for the time that model remains
- A team wants delivery steered by a plain-language instruction on gpt-4o-mini-tts before its shutdown
What these rows mean
- Audio Realism Bench, 4 August 2026: Design Arena's blind listening test, where native speakers hear two recordings of the same line and pick the one that sounds more like a real person. GPT-4o Mini TTS is the OpenAI model that runs on the speech endpoint. OpenAI's tts-1 and tts-1-hd weren't in these results.
- Model status: OpenAI announced the deprecations on 1 October 2026.
- Price per million characters: Bland Speech costs $0.04 per 1,000 characters. OpenAI charges $15 per million characters for tts-1 and $30 for tts-1-hd, and prices gpt-4o-mini-tts at $0.60 per million text input tokens and $12 per million audio output tokens.
- Phone audio: Mu-law is the 8-bit format US phone networks and Twilio carry. OpenAI's Realtime API session settings accept audio/pcmu, which is mu-law.
- Voice cloning: OpenAI limits custom voices made from recordings to eligible customers, with samples of 60 seconds or less and a recorded consent phrase.
How do Bland Speech and OpenAI TTS differ?
OpenAI TTS is a speech endpoint with three models that have a shutdown date, while Bland Speech is one current model behind an endpoint that accepts the same request shape.
Bland Speech overview
Bland Speech is the text to speech model inside Bland, a voice AI provider that runs its own model and speech rather than wrapping a third party's. It is sold on its own as an API and a studio on the Bland Speech page.
A developer sends text over HTTP to POST /v2/tts and gets raw audio or a WAV file back, or streams tokens from a language model into a WebSocket at WSS /v2/tts/ws, where a new context immediately interrupts the turn that is playing. That interruption is called barge-in, and it is how a voice agent stops talking when a caller cuts in. POST /v2/audio/speech implements OpenAI's audio speech contract. The text to speech API page has the endpoints and code.
Pros
- Bland Speech v3 ranked first of 13 AI voices on Audio Realism Bench on 4 August 2026
- An endpoint that implements OpenAI's audio speech contract
- One rate, $0.04 per 1,000 characters, in the studio and the API
- HTTP and WebSocket streaming, with barge-in in the protocol
- Mu-law at 8 kHz and PCM from 8 to 48 kHz
- Performance tags for how a line is read
Cons
- More than a hobby project needs, because Bland Speech is built for production calls
OpenAI TTS overview
OpenAI's speech endpoint, /v1/audio/speech, serves gpt-4o-mini-tts, tts-1, and tts-1-hd, and OpenAI has deprecated all three, with shutdown on 6 January 2027. OpenAI names gpt-realtime-2.1-mini on the Realtime API as the migration target, and says it isn't a supported model for the speech endpoint.
On the speech endpoint, audio streams over HTTP with chunked transfer encoding. WebSocket, a connection that stays open so text and audio flow both ways, and WebRTC sessions belong to the Realtime API.
Pros
- tts-1 at $15 per million characters and tts-1-hd at $30
- A plain-language instructions field on gpt-4o-mini-tts for accent, tone, and speed
- A Realtime API with WebSocket and WebRTC sessions, and mu-law output in its session settings
- SOC 2 Type 2 for its API, with a DPA and a BAA offered
Cons
- tts-1, tts-1-hd, and gpt-4o-mini-tts shut down on 6 January 2027
- The replacement model isn't supported on the speech endpoint, so it is a different API workflow
- GPT-4o Mini TTS ranked 9th of 13 AI voices on Audio Realism Bench on 4 August 2026
- The speech endpoint returns mp3, opus, aac, flac, wav, and pcm at 24 kHz, with no mu-law
- Custom voices from recordings are limited to eligible customers
What changes when an OpenAI client points at Bland Speech?
Three settings change and a few behaviors differ. Bland says a team changes the base URL, the API key, and the model to route speech through Bland, and the current model id is btts-3. OpenAI model ids such as tts-1 and gpt-4o-mini-tts are not aliased, so they have to be replaced.
Bland accepts a speed of 1.0 only, accepts the instructions field and ignores it, returns an error if a request asks for the sse stream format, and takes at most 4,096 characters of input. Formats are mp3, opus, aac, flac, wav, and pcm, with pcm at 24 kHz. A team that steered delivery with instructions moves that steering into Bland's inline tags, such as [say warmly], which tell the model how to read the words after them.
How do the two handle phone audio and latency?
Bland publishes 199ms to first audio on a streamed request, and time to first audio is the gap between sending text and hearing the first sound. OpenAI says tts-1 provides lower latency than tts-1-hd.
For a phone line, the format matters as much as the speed. Bland Speech returns mu-law at 8 kHz or PCM at 8,000, 16,000, 24,000, 44,100, or 48,000 Hz. OpenAI's speech endpoint returns mp3, opus, aac, flac, wav, and pcm at 24 kHz, while OpenAI's Realtime API accepts audio/pcmu, its name for mu-law, in session settings.
How do the two handle voice cloning and security?
Bland Speech makes an instant clone from 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. Every clone is private to the customer's organization.
OpenAI limits custom voices made from recordings to eligible customers, caps them at 20 per organization, and wants a sample of at least five seconds of speech, up to 60 seconds, plus a consent recording of an approved phrase. OpenAI's usage policies also require telling end users that a voice is AI-generated.
Bland is SOC 2 Type II audited, HIPAA compliant with a signed BAA, and FedRAMP 20x Class C certified at Moderate, granted on 17 November 2026, and Bland's trust page says speech synthesis runs inside the FedRAMP boundary. Enterprise customers get dedicated orchestration and dedicated GPUs run by Bland. OpenAI says it has completed a SOC 2 Type 2 examination for its API and offers a DPA and a BAA.
How much do Bland Speech and OpenAI TTS cost?
OpenAI's tts-1 and tts-1-hd cost less per character than Bland Speech, at $15 and $30 per million against $40, and both models shut down on 6 January 2027. After that, a team on OpenAI prices speech per audio token on the Realtime API, while Bland Speech stays one per-character rate.
Bland Speech pricing
Bland Speech costs $0.04 per 1,000 characters, directed or not, in the studio or the API, billed against a balance that doesn't expire. New accounts get 50,000 characters free. Bland's own examples are a 60-second note at $0.04, a 10-minute narration at $0.36, and a million characters at $40.
Enterprise pricing is custom, set with Bland's sales team.
OpenAI TTS pricing
OpenAI charges $15.00 per million characters for tts-1 and $30.00 for tts-1-hd. gpt-4o-mini-tts is priced per token, at $0.60 per million text input tokens and $12.00 per million audio output tokens. The replacement, gpt-realtime-2.1-mini, is priced at $10.00 per million audio input tokens, $20.00 per million audio output tokens, and $0.30 per million cached tokens.
What does a month of speech cost on each?
At per-character rates, 100,000 characters is $4 on Bland Speech, $1.50 on tts-1, and $3 on tts-1-hd. A million characters is $40, $15, and $30, and ten million characters is $400, $150, and $300.
tts-1 is the cheaper way to buy speech until 6 January 2027, and that price doesn't survive the shutdown. A team budgeting for 2027 is comparing Bland Speech's flat rate with a token-priced Realtime model.
Which sounds more human, Bland Speech or OpenAI TTS?
Bland Speech v3 scored above both OpenAI models Design Arena entered in the blind tests it published on 4 August 2026. In Design Arena's Audio Realism Bench results of 4 August 2026, Bland Speech v3 ranked first of 13 AI voices at 1365, OpenAI's GPT Realtime 2 ranked 4th at 1119, GPT-4o Mini TTS ranked 9th at 1011, and the human recording scored 1500.
- #1BlandBland Speech v31365
- 2ndMicrosoftMAI-Voice-21224
- 3rdxAIGrok TTS1156
- 4thOpenAIGPT Realtime 21119
- 5thGoogleGemini 2.5 Pro TTS1109
Ranks are among 13 AI voices, and OpenAI's own voice is on the list. Real human recordings scored 1500, at the right edge of this scale. Bars start at 600, not zero. Source: Design Arena's published results.
Scores are Elo ratings, the system chess uses, where a model gains more for beating a higher-rated one. GPT-4o Mini TTS runs on OpenAI's speech endpoint, and GPT Realtime 2 is a Realtime model. tts-1, tts-1-hd, and gpt-realtime-2.1-mini, the replacement OpenAI names, weren't in the results. Design Arena's published results explain how the votes were collected.
Why is Bland Speech the best OpenAI TTS alternative?
Bland Speech is the better OpenAI TTS alternative because it is a current model with a request shape an OpenAI integration already uses, and because Bland Speech v3 ranked first among AI voices on Audio Realism Bench on 4 August 2026. The shutdown date turns a someday migration into a scheduled one, and a matching request shape keeps the code change small.
The migration guide walks through the mapping, and the OpenAI-compatible text to speech APIs list shows other providers that accept the same contract. Bland's comparison of text to speech APIs covers more vendors, and the head-to-heads hub lists every pairing.
How was this comparison made?
OpenAI details were captured from OpenAI's developer guides, API reference, pricing, deprecations table, and security pages, and voice scores come from Design Arena's published results of 4 August 2026.
Frequently asked questions
Is Bland Speech better than OpenAI TTS?
For phone calls and voice agents, Bland Speech is the better choice. Bland Speech v3 ranked first of 13 AI voices on Audio Realism Bench on 4 August 2026 and OpenAI's GPT-4o Mini TTS ranked 9th. OpenAI suits a team building a whole conversation on its Realtime API.
Is OpenAI shutting down its text to speech models?
Yes. OpenAI has deprecated tts-1, tts-1-hd, and gpt-4o-mini-tts, and shuts them down on 6 January 2027. OpenAI names gpt-realtime-2.1-mini on the Realtime API as the replacement and says it isn't supported on the speech endpoint.
Can Bland Speech replace an OpenAI text to speech integration?
Yes. POST /v2/audio/speech implements OpenAI's audio speech contract, so a team changes the base URL, the API key, and the model, using btts-3 in place of tts-1 or gpt-4o-mini-tts. Speed accepts only 1.0, and the instructions field is accepted and ignored.
How much does Bland Speech cost?
Bland Speech costs $0.04 per 1,000 characters, which is $40 per million, in the studio or the API. New accounts get 50,000 characters free, and credits do not expire. Enterprise pricing is custom, set with Bland's sales team.
Does OpenAI TTS sound as human as Bland Speech?
In Design Arena's results of 4 August 2026, OpenAI's GPT-4o Mini TTS ranked 9th of 13 AI voices at 1011 and GPT Realtime 2 ranked 4th at 1119, against 1365 for Bland Speech v3. tts-1 and tts-1-hd weren't in those results.
Can Bland Speech clone a voice?
Yes. An instant clone takes one clean sample of about 10 seconds, and a professional clone is fine-tuned on 30 minutes or more of audio after a consent and rights review. OpenAI limits custom voices made from recordings to eligible customers.
Does Bland Speech work on phone calls?
Yes. Bland built Bland Speech for phone calls, and it returns mu-law at 8 kHz as well as PCM from 8 to 48 kHz. OpenAI's speech endpoint returns mp3, opus, aac, flac, wav, and pcm at 24 kHz.
