OpenAI Is Killing Its TTS Models in 2027: 7 Free Alternatives

1. What exactly is shutting down (and when)

OpenAI's deprecations page lists four model IDs with the same cutoff date:

• tts-1 — removed January 6, 2027
• tts-1-hd — removed January 6, 2027
• gpt-4o-mini-tts-2025-03-20 — removed January 6, 2027
• gpt-4o-mini-tts-2025-12-15 — removed January 6, 2027

These were the simple, beloved speech models: text in, audio out, through a single v1/audio/speech API call. Developers picked one of 11 preset voices (alloy, echo, fable, nova, onyx, shimmer, and others) and got reliable narration back. Apps that turn articles into audio, podcast tools, and accessibility services were built on exactly this simplicity.

Azure has its own, earlier schedule: some gpt-4o-mini-tts versions retire October 15, 2026, with Azure's whisper, tts, and tts-hd deployments following on December 15, 2026. If your app runs on Azure OpenAI, check your deployment this week — the deadlines there are days away, not months.

---

2. Why the "replacement" is frustrating everyone

OpenAI names gpt-realtime-2.1-mini as the replacement for all four models. The problem: it is not a drop-in replacement at all.

• It is a Realtime API model, which means WebSocket or SIP sessions instead of one simple HTTP request. One developer put it bluntly: "a few lines of code turn into 250+ lines now to handle an async WebSocket connection."
• Its own documentation lists v1/audio/speech as "Not supported" — the exact endpoint everyone is migrating away from.
• The voice lineup does not match: familiar voices like fable, nova, and onyx are not in the current Realtime list.

The backlash was immediate. The founder of Listen Later — an app that turns articles into audio for hundreds of long-time listeners, many of them visually impaired — publicly asked OpenAI to reconsider, arguing that "a familiar narrator becomes part of someone's routine" and a better model cannot replace years of familiarity. Another developer noted the new models are "conversational" by design, while the old TTS models did one job reliably: text in, speech out.

One community suggestion is worth knowing: gpt-audio-1.5 can return audio through a single /v1/chat/completions request without WebSockets — a much smaller migration if you want to stay in OpenAI's ecosystem. But the writing on the wall is clear: OpenAI is moving away from dedicated, simple TTS models toward multimodal realtime models.

---

3. What to do if you rely on these models

If you are a developer: inventory every place your code calls tts-1, tts-1-hd, or gpt-4o-mini-tts now. Test the migration path (Realtime API sessions or gpt-audio-1.5) well before January 6 — deprecation is advance notice, not an immediate outage. Budget real engineering time; the WebSocket rewrite is not trivial.

If you use an app built on OpenAI TTS (article readers, audiobook tools, voice features in apps): you do not need to do anything yet. Your app's developer handles the migration. But if your favorite app starts sounding different in 2027, now you know why.

If you just want free text-to-speech for yourself — voiceovers, reading documents aloud, narrating a video — you were never dependent on these models, and the free options below are simpler than any API migration.

---

4. Seven free alternatives that work right now

1. TextToSpeechH — best overall free replacement

The zero-friction option. Paste up to 10,000 words, pick a voice and emotion, adjust speed and pitch, and download a clean MP3 — no account, no credit card, no API key. It also imports PDF, DOCX, and TXT files directly, which covers the "turn this document into audio" use case the OpenAI models were popular for. If the shutdown news sent you hunting for a replacement, try it free here.

2. ElevenLabs — best voice quality on a free tier

The quality benchmark in AI voices. The free plan gives you 10,000 credits a month (roughly 10 minutes of audio) with MP3 download — enough to test whether the voices beat everything else. Note: the free tier is non-commercial; paid plans add commercial rights plus voice cloning. See our ElevenLabs v4 free guide for what the newest voices can do.

3. TTSMaker — most generous free allowance

A genuinely free workhorse: 20,000 characters per week plus unlimited access to some voices, with commercial use allowed and no credit needed. The interface is not fancy, but the allowance is generous for a free tool.

4. Microsoft Edge Read Aloud — best completely-free listening

Already on your computer if you use Windows. Open any web page or PDF in Edge, hit Ctrl+Shift+U, and it reads aloud with surprisingly decent neural voices. Unlimited, free, zero setup. The voices are not as expressive as dedicated TTS tools, but for listening to articles it is the fastest path that exists.

5. Google Cloud Text-to-Speech — best free tier for developers

If you are migrating an app off OpenAI's API, this is the closest thing to a familiar developer experience: up to 4 million characters per month on the free tier, a real REST API, and WaveNet voices. You will need a Google Cloud account, but the free allowance covers serious testing.

6. Cartesia — best for real-time voice

Built for real-time voice AI with very low latency. The free tier includes 20,000 credits a month, and the $5/month Pro plan unlocks instant voice cloning. A strong pick if your use case was real-time or conversational speech.

7. Murf — best studio-style free trial

A polished studio-style tool aimed at video and e-learning voiceovers. The free plan gives you about 10 minutes to test (watermarked, no download); paid plans start around $29/month. Worth trying if your workflow is scripted narration with a timeline.

---

The bottom line

OpenAI's TTS shutdown is a reminder that building on someone else's API means living on their schedule. The good news: you do not need an API at all for most text-to-speech jobs. For everyday use — documents, scripts, voiceovers — a free browser tool like TextToSpeechH does the job in two minutes, with no migration, no WebSocket code, and no January deadline hanging over it.

And if this is the second shutdown notice you have read this month — Play.ht went the same way — the pattern is worth noticing: free, simple TTS keeps moving toward the open web, not locked-in APIs.

Frequently Asked Questions