Guide · Voice
Cheapest Text-to-Speech APIs in 2026
The cheapest text-to-speech APIs in 2026 are the standard voices from the large cloud providers, with Amazon Polly and Google Cloud Text-to-Speech sitting at the bottom of the cost-per-million-characters table. They are the default answer whenever price per character is the number you are optimizing, and for high-volume, functional speech they are hard to beat.
But cheapest is only the right target for some jobs. The neural voices from ElevenLabs and Cartesia cost meaningfully more per million characters and sound noticeably more human, which changes the math the moment your audio is customer-facing. This guide ranks TTS APIs on the real cost per million characters we track, then explains where paying more actually earns its keep.
Cheapest APIs by cost per million characters
Ranked by the real cost per million characters in our database, with standard and neural tiers shown separately because they are not the same product. Prices update live, so the order reflects current rates.
Unreal Speech is a low-cost, high-throughput text-to-speech API that positions itself as one of the cheapest TTS providers for developers.
Rime AI is a developer-focused TTS API built for voice agents, offering ultra-low-latency naturalistic voices via its Mist and Arcana models.
Resemble AI is a voice-cloning platform and API offering custom synthetic voices, real-time text-to-speech, and an AI voice changer on pay-as-you-go pricing.
How to read cost per million characters
Cost per million characters is the only honest way to compare TTS pricing, because it strips out the packaging. Vendors quote by characters, by credits, by minutes of audio and by monthly bundles, and those units hide real differences of several times over. Normalize everything to one million characters and the table reorders itself: the standard cloud voices land far below the premium neural services, and 'unlimited' consumer plans usually turn out to have per-character economics buried in a fair-use clause.
Two adjustments matter once you have the base rate. First, tier: the same provider often sells a cheap standard voice and a pricier neural one, so compare like for like. Second, real usage rather than list price: free monthly characters, volume discounts and whether you pay for retries or failed calls all move the effective cost. Our numbers track the rate you actually pay at volume, not the headline figure.
Cheap vs premium: when quality is worth the cost
Go cheap when the audio is functional and high-volume. Amazon Polly and Google Cloud TTS are the right call for IVR menus, notifications, internal tooling, accessibility read-back and anything where the listener wants information, not performance. At scale the cost gap is large enough that paying for premium voices here is simply waste.
Pay more when the voice is the product. ElevenLabs and Cartesia earn their higher cost per million characters when audio is customer-facing and the delivery carries the brand: narration, characters, ads, agents people talk to, anything where a flat or robotic read costs you trust or attention. Cartesia additionally competes on latency, which matters for real-time conversational agents where a slow response breaks the illusion. The right move is often a split: cheap standard voices for the functional bulk, a premium API only for the surfaces that people judge you on.
Frequently asked
What is the cheapest text-to-speech API?+
For raw cost per million characters, the standard voices from Amazon Polly and Google Cloud Text-to-Speech are consistently at the bottom of our table. They are the cheapest way to turn large volumes of text into acceptable speech, as long as you do not need the most natural-sounding delivery.
Why are ElevenLabs and Cartesia more expensive?+
You are paying for voice quality and, in Cartesia's case, low latency. Their neural models sound closer to a human performer than the standard cloud voices, which is worth the premium for customer-facing audio and worth nothing for an IVR menu. The cost-per-million-character gap is the price of that realism.
Should I pick one API or mix them?+
Mixing is usually the cheapest good answer. Route high-volume functional speech to a low-cost standard voice and reserve a premium API for the audio your audience actually judges. It keeps your average cost per million characters down while protecting the surfaces where a robotic voice would cost you.