Short answer: The genuinely free ElevenLabs alternatives in 2026 fall into two buckets. Open-source models — Kokoro (Apache 2.0), Chatterbox (MIT), and the newer Qwen3-TTS (Apache 2.0) — are free to run and now rival ElevenLabs in blind listening tests, but you have to host them or use a cheap API. Freemium SaaS — Murf, Fliki, and ElevenLabs itself — give you a limited free monthly quota, usually with a watermark or no commercial rights. The catch most lists skip: free tools almost always paywall the file export, so you can listen but not download without paying.
What "free" actually means here
There are three different things people mean by "free TTS," and they have very different catches:
- Free to run (open source) — zero cost, unlimited use, but you provide the compute.
- Free tier (freemium SaaS) — a small monthly quota, often watermarked or non-commercial.
- Free to preview, pay to export — you can generate and listen, but downloading the audio file costs a subscription. The most common — and most frustrating — pattern.
Free open-source models (genuinely free, you host them)
| Model | License | Voice cloning | Notes |
|---|---|---|---|
| Kokoro (82M) | Apache 2.0 | No (fixed voices) | Tiny, fast, runs in a browser; still the easiest free option |
| Chatterbox V3 / Turbo | MIT | Yes (5-sec clone) | Multilingual V3 (Jun 2026) covers 23 languages; outputs carry Resemble's PerTh watermark |
| Qwen3-TTS | Apache 2.0 | Yes (3-sec zero-shot) | Jan 2026, from Alibaba; 10 languages, streaming with ~100ms first audio; the strongest permissive newcomer |
| Kyutai Pocket TTS | MIT code; gated weights | Yes (one sample) | Runs at 6× realtime on 2 laptop CPU cores — no GPU needed; 6 European languages |
| Fish Audio S2 Pro | Fish Audio Research License | Yes | Best open-weights quality on the arena — but commercial use now requires a paid license |
| Higgs Audio v3 (4B) | Non-commercial + creator grant | Yes (zero-shot) | Free for monetized videos/podcasts if you credit Boson AI; 100+ languages |
| F5-TTS | CC-BY-NC | Yes | Non-commercial license — personal use only, NOT for a business |
| StyleTTS2 | MIT | Yes | Commercial-safe but aging; avoid XTTS v2 for business — its non-commercial CPML can't be upgraded since Coqui shut down |
The 2026 headline: open source caught up, then reshuffled. Fish Audio's S2 Pro now tops the open-weights side of the Artificial Analysis Speech Arena, and ElevenLabs' flagship no longer leads the overall board. Resemble's own blind tests have listeners preferring Chatterbox over ElevenLabs about 64% of the time — the arena is less generous to it, but either way "free" no longer means "worse." The trade-off is unchanged: you host the model or use a cheap API. A hosted Kokoro endpoint costs about $0.62 per 1M characters on DeepInfra — effectively free for normal use. Watch the licenses, though: two models on this list (Fish and F5) are free to run but not free to sell with.
Free SaaS tiers (no setup, but limited)
- ElevenLabs Free — 10,000 credits/month, but no commercial license; commercial rights start on the $6/mo Starter plan.
- Murf Free — 10 minutes of generation to trial the studio; no downloads and no commercial rights.
- Fliki Free — 3 credits/month with a 1-minute export cap, watermarked; good for testing text-to-video voice.
- Play.ht (PlayAI) — small free quota with attribution. LOVO now leads with a 14-day Pro trial rather than a perpetual free tier.
These are real but capped, and the free output is usually watermarked or non-commercial — fine for testing, not for shipping.
Enjoying this? Get one like it in your inbox each morning.
one email a day · unsubscribe in two clicks · no third-party tracking
The catch nobody mentions: the export paywall
Here's the gotcha that turns "free" into a subscription: the popular read-aloud tools — Speechify, NaturalReader, the ElevenReader app — let you listen for free but paywall downloading the MP3. If your goal is "turn this article/PDF/script into an audio file I can keep," a free reader won't give you the file.
That's the one job Stork's Article-to-Audio tool does without a subscription: paste text or upload a PDF, download the MP3, pay once. No monthly fee, no watermark on the file.
So which free option should you use?
- Just want to listen? Any free SaaS tier works.
- Want to download a file once? Use a pay-once exporter, not a "free" reader that paywalls export.
- Building something / generating at volume? Self-host Kokoro, Chatterbox, or Qwen3-TTS, or use a cheap hosted endpoint (from ~$0.62 per 1M characters).
- Need commercial rights? Stick to the Apache/MIT models — Kokoro, Chatterbox, Qwen3-TTS, StyleTTS2. Avoid F5-TTS and XTTS v2 (non-commercial), Fish S2 without a paid license, and watermarked free SaaS tiers.
FAQ
Is there a 100% free ElevenLabs alternative? Yes — Kokoro (Apache 2.0), Chatterbox (MIT), and Qwen3-TTS (Apache 2.0) are free to run with no usage cost, and in Resemble's blind tests listeners preferred Chatterbox to ElevenLabs. You just have to host them or use a cheap API.
Can I use free TTS commercially? Only some. The Apache/MIT models — Kokoro, Chatterbox, Qwen3-TTS, StyleTTS2 — allow commercial use. Watch the traps: Fish Audio's S2 weights now require a paid commercial license, F5-TTS is non-commercial, and XTTS v2's non-commercial license can no longer be upgraded because Coqui shut down.
Why won't free tools let me download the audio? Many readers (Speechify, NaturalReader, ElevenReader) monetize by paywalling the export. The generation is free; the file costs money. A pay-once exporter avoids the subscription.
Is free TTS worse than ElevenLabs? Not automatically, not anymore. Resemble's blind tests prefer Chatterbox, and on the Artificial Analysis arena several models now outrank ElevenLabs' flagship — though ElevenLabs still beats most open models there. For a full breakdown by use case, see our best ElevenLabs alternatives guide.