development

Best Text-to-Speech APIs

The best text-to-speech APIs in 2026 compared for developers — ElevenLabs, Cartesia, Google Gemini TTS, Inworld, SpeechifyAI, Deepgram, Rime, OpenAI, Hume, Azure and Amazon Polly, with independent quality and latency rankings, streaming protocols, per-character and per-token pricing, voice cloning, SSML control and decisive picks for real-time voice agents, high-volume synthesis and self-hosting.

Updated 6 November 2026

Quick answer: For the best voice quality, the ElevenLabs API leads: Eleven v4, launched on 28 September, was first on the Artificial Analysis Text-to-Speech leaderboard on 29 September 2026, covers 90+ languages and comes with a real-time v4 Turbo variant at around 150ms to first speech. For real-time voice agents, Cartesia Sonic 3.6 (second on Artificial Analysis that day) and Inworld Realtime TTS-2 (fifth) pair top-ten quality with low latency, while SpeechifyAI Simba 3.2 (seventh) was the fastest model on Voice Arena’s latency board in October 2026, at 123ms, and costs $6–10 per million characters, though it speaks English only. For cheap top-tier synthesis across 100+ languages, Google Gemini 3.8 Flash-Lite TTS costs $6 per million audio tokens — about $0.09 for a 10-minute narration — until the price doubles on 1 January 2027. Deepgram Flux TTS is the natural pick for teams already running Deepgram speech-to-text, and Rime the specialist for phone agents that need conversational prosody. For emotion-critical delivery, Hume Octave 2 leads on explicit emotional control; for enterprise compliance and language breadth, Microsoft Azure AI Speech covers 140+ languages with a HIPAA and SOC story; for high-volume AWS-native workloads, Amazon Polly is the cost-predictable workhorse. The one thing to decide first: whether you need a standalone TTS API at all, because bundled speech-to-speech models such as OpenAI gpt-realtime-2.1 now collapse the whole voice pipeline into one call for many agent builds.

This guide ranks the APIs you call programmatically to turn text into speech inside a product — measured on latency, streaming, price and control, not on a web dashboard. If you want a creator or consumer tool with a web interface, a voice library and produced voice-overs, that is a different job: see our best AI voice generator guide, which covers ElevenLabs, Murf, the Speechify reading app and the rest as tools rather than endpoints. For the inverse problem — turning speech into text — see best AI for transcription. For cloning-specific picks and the legal landscape around it, see best AI voice clone.

The current state of text-to-speech APIs

The text-to-speech API market has stopped competing on raw naturalness and started competing on latency, price and control.

The market backing this is large and growing. The text-to-speech market sits near $4.36 billion in 2026, up from $3.87 billion in 2025, and is forecast to reach roughly $7.9 billion by 2031 at a compound annual growth rate around 12.7% (Mordor Intelligence). Forecasts for the adjacent AI voice generator category run considerably higher, so treat any single headline number as one analyst’s definition of the market rather than a settled figure. The growth driver is no longer audiobooks or e-learning voice-overs — it is real-time voice agents, where speech is generated live inside a phone call or assistant and every millisecond of delay is audible.

Six shifts define the current moment.

First, the quality leaderboard turned over in six weeks, from mid-August to late September 2026. In that time Cartesia shipped Sonic 3.6, Google shipped Gemini 3.8 Flash TTS and Flash-Lite TTS, and ElevenLabs shipped Eleven v4. On 29 September 2026 the Artificial Analysis leaderboard of 92 models, ranked by blind listener votes, read: Eleven v4, Cartesia Sonic 3.6, Gemini 3.8 Flash TTS, Alibaba’s Qwen-Audio-3.0-TTS-Plus, Inworld Realtime TTS-2, Gemini 3.8 Flash-Lite TTS and SpeechifyAI Simba 3.2. Sonic 3.5 and Eleven v3, the flagships this guide ranked in August, now sit outside the top ten.

Second, latency became the battleground. Vendors now quote time-to-first-audio in tens of milliseconds: Inworld claims 25ms time-to-first-byte for Realtime TTS-2 Flash, Deepgram 80ms for Flux TTS and Cartesia sub-90ms for Sonic 3.6. Independent measurement runs higher: in October 2026 Voice Arena measured a 123ms median for SpeechifyAI Simba 3.2, the fastest of the 14 models on its board, and 340.5ms for Cartesia Sonic 3.6. Sub-300ms time-to-first-audio is the threshold below which turn-taking in a conversation feels natural.

Third, the market split into three layers. Expressive production models (ElevenLabs Eleven v4, Hume Octave 2) optimise for quality and emotional range; real-time agent models (Cartesia Sonic, Inworld Realtime TTS-2, SpeechifyAI Simba, Deepgram Flux TTS, Rime, ElevenLabs v4 Turbo) optimise for streaming latency; and high-volume cloud models (Google, Amazon Polly, Azure) optimise for cost and compliance at scale. Choosing well means choosing the layer before the vendor.

Fourth, billing moved from characters to tokens on the newest models. Traditional TTS APIs charge per character — $4 to $80 per million is the common band. The newest models from Google and OpenAI bill on audio tokens instead: Google Gemini 3.8 Flash TTS costs $9 and Flash-Lite TTS $6 per million output audio tokens until 31 December 2026, doubling to $18 and $12 from 1 January 2027. Token billing means you should estimate cost from audio duration, not text length.

Fifth, streaming and WebSocket delivery became standard. Every agent-grade API now supports chunked HTTP or WebSocket streaming so audio starts playing before the full sentence is synthesised. File-return endpoints still exist for batch narration, but they are no longer competitive for conversational use.

Sixth, the speech-to-speech shift is reshaping whether you need a TTS API at all. OpenAI gpt-realtime-2.1 and SpaceXSI Grok Voice bundle transcription, reasoning and speech into a single model, removing the separate text-to-speech step for some agent builds (TestingCatalog). A standalone TTS API still wins when you want a specific branded voice, cheaper high-volume synthesis, or a best-of-breed pipeline you control end to end.

One market exit sits behind all of this: PlayHT shut down permanently on 31 December 2025 after Meta acquihired the PlayAI team, deleting all accounts, voice clones and audio with no migration path. If you are choosing an API today, choose one with a durable business behind it.

The three types of text-to-speech API

Match the layer to the job before comparing vendors within it.

Most production voice systems in 2026 use more than one layer: a low-latency streaming model for the live conversational turn, and a cheaper cloud or self-hosted model for high-volume, latency-tolerant synthesis such as batch notifications.

Top text-to-speech APIs ranked

Ranked by overall suitability for a team building a new voice product today, weighting voice quality, latency, streaming support, price and language coverage. Quality positions are from the Artificial Analysis leaderboard on 29 September 2026. Latency is vendor-reported unless marked “Voice Arena”, which is an independent median measured by Voice Arena in October 2026; verify against your own region and payload before committing.

Also worth knowing: Alibaba’s Qwen-Audio-3.0-TTS-Plus was fourth on the Artificial Analysis leaderboard on 29 September 2026 and is served through Alibaba Cloud; MiniMax Speech 2.8 serves 40+ languages with strong expressiveness at around $100 per million characters (HD); LMNT and Neuphonic are low-latency developer options priced at roughly $43.60 and $17.60 per million characters respectively; and for self-hosting the open-weight Kokoro-82M and Chatterbox models run locally at no per-character cost.

The best text-to-speech APIs compared

Ordered by current suitability for a new voice build, not by company size. Every price is list; verify against each vendor’s live pricing page before you commit spend.

1. ElevenLabs API — best voice quality and cloning

Type: Expressive and real-time Flagship: Eleven v4 (produced) and Eleven v4 Turbo (real-time) Pricing: API list price $0.08 per 1,000 characters for Eleven v4 and $0.04 for v4 Turbo, both 72% off until 12 October 2026 ($0.022 and $0.011); subscription plans from $6/month Starter, with a free plan of 10,000 credits a month (ElevenLabs) Key features: 90+ languages, 10,000-character single generations, natural-language direction and inline audio tags, bidirectional streaming on v4 Turbo, instant and professional voice cloning with verified consent

The ElevenLabs API is the quality benchmark. Eleven v4, launched on 28 September 2026 alongside the low-latency Eleven v4 Turbo, was first of 92 models on the Artificial Analysis leaderboard on 29 September 2026, with an Elo of 1,319, 43 points clear of second place. Both models cover 90+ languages, and a voice recorded in one language can speak another while keeping the speaker’s identity (TechCrunch). The model split is the key API decision: Eleven v4 is the expressive flagship, while v4 Turbo is the agent model, which ElevenLabs puts at around 100ms median inference latency and around 150ms median time to first speech. Pick v4 for produced audio; pick v4 Turbo for conversation.

Why it wins: The highest-rated voice quality on the independent leaderboard on 29 September 2026, the deepest cloning options and the broadest language coverage of the specialist APIs, with a production and a real-time model on one account.

Limitations: It is the most expensive API here at list price: $80 per million characters for v4 and $40 for v4 Turbo once the launch discount ends. The credit-based subscription billing draws the most developer complaints — failed generations and retries consume credits, so real costs often exceed the advertised per-character estimate. ElevenLabs is also a defendant in the BIPA class actions filed in Chicago federal court between 11 and 13 May 2026 over alleged unauthorised voiceprint training — though this is not an ElevenLabs-specific risk: the same coordinated set of nine suits names Adobe, Alphabet, Amazon, Apple, Meta, Microsoft, NVIDIA and Samsung, so Google, Azure and Polly sit under the same cloud.

Best for: Products where voice quality directly affects revenue — branded assistants, characters, produced narration — and any team that needs both a production and a real-time model from one vendor.

2. Cartesia — high-quality, low-latency voice agents

Type: Real-time Flagship: Sonic 3.6 (dated snapshot 27 August 2026) Pricing: Credit plans — Pro $5/month (100K credits), Startup $49 (1.25M), Scale $299 (8M) — with standard TTS at about one credit per character, roughly $37–50 per million characters depending on plan; free plan available (Cartesia) Key features: State-space (SSM) architecture, WebSocket and SSE streaming, instant and professional voice cloning, 44 languages, stable dated model snapshots

Cartesia built its Sonic models for real-time agents, and Sonic 3.6 pairs that with top-tier quality: it was second on the Artificial Analysis leaderboard on 29 September 2026, at 1,276 Elo. Cartesia quotes sub-90ms model latency; Voice Arena’s independent measurement in October 2026 put Sonic 3.6 at a 340.5ms median time-to-first-audio and Sonic 3.5 at 250ms, so treat the vendor figure as the model’s own contribution, not the whole round trip. Sonic 3.6 is backwards compatible with Sonic 3.5 and adds Odia and Urdu, for 44 languages.

Why it wins: Second-best voice quality on the independent leaderboard on 29 September 2026, a streaming API built for the “bring your own speech-to-text and language model” pattern most custom voice agents use, and dated snapshots you can pin in production.

Limitations: Cartesia is built for developers, not content teams — there is no polished editing interface or large stock-voice catalogue. Independent latency measurements for Sonic 3.6 run well above the vendor figure, and professional cloning is billed at a higher credit rate.

Best for: Developers building real-time voice agents, IVR systems and live applications that need both leaderboard-level quality and streaming latency.

3. Google Gemini TTS — cheapest top-tier quality at scale

Type: High-volume Flagship: Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS (Gemini API and AI Studio), with Chirp 3 HD in Cloud Text-to-Speech Pricing: Gemini 3.8 Flash TTS $0.50 per million input tokens and $9 per million output audio tokens; Flash-Lite TTS $0.50 and $6 — both until 31 December 2026, then $1 and $18 (Flash) or $12 (Flash-Lite) from 1 January 2027; batch and flex at half price; a free tier on the Gemini API (Google). Chirp 3 HD is $30 per million characters in Cloud Text-to-Speech (Google Cloud) Key features: 100+ languages, 2,000+ voices, voice design from a written prompt, voice replication from a 30-second sample with recorded consent, SynthID watermarking on all output

Google put Gemini 3.8 Flash TTS and Flash-Lite TTS into general availability on the Gemini API and AI Studio on 23 September 2026, and on 29 September they were third and sixth on the Artificial Analysis leaderboard. Google aims Flash at audiobooks, podcasts and games and Flash-Lite at high-volume dubbing and voice agents. On price, Google puts Flash-Lite at about $0.0015 per 10 seconds of audio — roughly $0.09 for a 10-minute narration — and Flash at about $0.0023 per 10 seconds, until the January price rise.

Why it wins: Top-ten quality at one of the lowest per-minute costs here, 100+ languages, a free tier for prototyping, and Google Cloud’s identity, logging and data-residency controls.

Limitations: Both prices double on 1 January 2027, so budget on the 2027 rates for anything long-running. Voice replication through AI Studio is not available in Illinois, Texas, the EEA, the UK, Switzerland or India, and every clip carries a SynthID watermark. Google does not publish latency for the 3.8 models; Voice Arena measured the older Gemini 3.1 Flash TTS at an 890ms median, so test before using it for tight turn-taking. Token billing means you must estimate from audio duration, not character count.

Best for: Cost-sensitive high-volume synthesis, multilingual narration and dubbing, and teams already building on Google Cloud.

4. Inworld — steerable real-time voice in 200+ languages

Type: Real-time Flagship: Realtime TTS-2 (quality) and Realtime TTS-2 Flash (latency and volume) Pricing: Realtime TTS-2 $25 per million characters on demand, falling to $12.50 on Growth and as low as $5 at enterprise scale; TTS-2 Flash $15 on demand, $7 on Growth and sub-$5 at enterprise; the on-demand tier includes up to 70 minutes of TTS free (Inworld) Key features: WebSocket streaming, 200+ languages (vendor figure), instant cross-lingual cloning from 5–15 seconds of audio, natural-language steering of tone, emotion, speed and volume, non-verbal cues such as laughs and sighs

Inworld Realtime TTS-2 was fifth on the Artificial Analysis leaderboard on 29 September 2026, and in October Voice Arena measured it as the second-fastest model on its latency board, at a 168.5ms median. Inworld quotes sub-100ms P99 time-to-first-byte for TTS-2 and 25ms for TTS-2 Flash, both measured server-side. You steer delivery with bracketed plain-language instructions such as “[Speak sadly]” rather than choosing a fixed voice, and Inworld bundles TTS, speech-to-text, a Realtime API and LLM routing behind one API and one bill. Its roots in game NPC dialogue make it a strong pick for interactive and gaming use.

Why it wins: Top-five quality, independently measured low latency and fine-grained natural-language steering, with the widest vendor-stated language list of the real-time APIs and aggressive per-character pricing at scale.

Limitations: The on-demand rate ($25 per million characters) is well above the scale price, so the headline $5 figure needs enterprise volume. The all-in-one platform is most valuable if you use several of its pieces.

Best for: Interactive and gaming voice, multilingual agents, and teams that want steerable real-time speech plus the rest of the voice pipeline from one account.

5. SpeechifyAI — low-latency, low-cost English voice

Type: Real-time Flagship: Simba 3.2 (English), with Simba 3.0 for German, Spanish, French, Italian and Brazilian Portuguese Pricing: Starter $10/month, then $10 per million characters; Pro $99/month at $8; Scale $499/month at $6; Enterprise custom; 500,000 free characters a month (SpeechifyAI) Key features: HTTP streaming including a timestamped stream endpoint, SSML and emotion control, zero-shot voice cloning from a 10–30 second sample on paid plans

SpeechifyAI is the developer API from Speechify, the company behind the Speechify reading app, running on its own Simba models. On 29 September 2026 Simba 3.2 was seventh of 92 on the Artificial Analysis leaderboard at 1,240 Elo, one point behind Gemini 3.8 Flash-Lite TTS and within its confidence interval, and in October Voice Arena measured it at a 123ms median time-to-first-audio, the fastest of the 14 models on its latency board. Its list price of $10 per million characters, falling to $6 on Scale, compares with $25 on demand for Inworld Realtime TTS-2 and roughly $37 or more for Cartesia.

Why it wins: Top-ten quality and the lowest independently measured latency, at a flat per-character price with a usable free allowance.

Limitations: Simba 3.2 is English-only — a request with a non-English voice returns an error — and Simba 3.0 covers five further languages, far narrower than ElevenLabs, Google or Inworld (SpeechifyAI docs). Voice cloning needs a paid plan, and voice agents are an enterprise product quoted per contract rather than part of the self-serve plans.

Best for: English-language voice agents and in-app speech where latency and per-character cost matter more than language breadth.

6. Deepgram — best for voice agents on the Deepgram stack

Type: Real-time Flagship: Flux TTS (launched 12 August 2026), with Aura-2 Pricing: Flux TTS $0.045 per 1,000 characters ($45 per million) with volume discounts; Aura-2 $30 per million; $200 free credit on pay-as-you-go (Deepgram) Key features: Output from as low as 80ms under production load, WebSocket streaming, conversation-aware delivery that carries context across turns, designed to read account numbers, drug names and amounts accurately, cloud, private-cloud or on-prem deployment

Deepgram built Flux TTS as the speaking half of the Flux conversational stack whose listening half — Flux speech-to-text with turn detection — is already a developer default for voice agents. Flux TTS treats the ongoing exchange rather than an isolated sentence as its context and accepts streaming LLM output, so speech starts before the language model finishes its reply. It launched with English voices, with more languages, voice cloning and finer emotional controls on Deepgram’s roadmap (audioXpress). Aura-2 remains available at $30 per million characters across 7 languages.

Why it wins: Low-latency streaming built for the agent loop, with the tightest integration path — one vendor, one SDK, one bill — for anyone already using Deepgram for transcription, plus on-prem deployment.

Limitations: Neither Flux TTS nor Aura-2 was in the Artificial Analysis top 15 on 29 September 2026, Flux TTS costs more per character than Aura-2, and there is no voice cloning yet. Language coverage is narrow next to the cloud providers.

Best for: Real-time voice agents and call-centre systems, especially teams already running Deepgram Flux or Nova for speech-to-text.

7. Rime — best for phone agents with conversational prosody

Type: Real-time Flagship: Coda (recommended for new builds), with Mist v2 and Arcana Pricing: Per-model, roughly $30–50 per million characters (Mist ~$30, Arcana ~$40, Coda ~$50), with free minutes on the Starter plan (Rime) Key features: 300+ voices, sub-200ms cloud latency (sub-100ms on-prem), mid-conversation code-switching that preserves voice identity, on-premise deployment

Rime is a text-to-speech API built for one job: making enterprise phone voice agents sound like people rather than IVR menus. Its models are tuned for authentic conversational prosody — the hesitations, emphasis and rhythm of real speech — rather than the smooth, even delivery of narration engines. Rime targets developers with 300+ voices and sub-200ms cloud latency, dropping to sub-100ms with on-premise deployment for teams that need speech to stay inside their own network. Its newer Coda model is the recommended default for new applications, with Mist v2 for high-performance general use.

Why it wins: Conversational-first voice quality plus an on-premise option, aimed precisely at the high-volume phone-agent use case where prosody and data control both matter.

Limitations: Rime is a specialist — it is not the tool for produced audiobooks or multilingual narration, and its language breadth trails the cloud providers. Model lineups move quickly, so pin a specific model version in production.

Best for: Enterprise phone and contact-centre voice agents that need natural conversational delivery and, optionally, on-premise deployment.

8. OpenAI — best for OpenAI-native pipelines and speech-to-speech

Type: Expressive and speech-to-speech Flagship: gpt-4o-mini-tts, tts-1 / tts-1-hd, and gpt-realtime-2.1 for speech-to-speech Pricing: tts-1 $15 and tts-1-hd $30 per million characters; gpt-4o-mini-tts token-billed at roughly $15 per million characters (about $0.015 per minute of audio); gpt-realtime-2.1 $32/$64 per million audio tokens in/out, and gpt-realtime-2.1-mini $10/$20 Key features: Streaming output, reasoning and tool use in the Realtime line, WebRTC, WebSocket and SIP connections, one account for text, speech and transcription

OpenAI covers text-to-speech from the same account as everything else you build on it. The tts-1 and tts-1-hd endpoints are the simple character-billed options at $15 and $30 per million characters. gpt-4o-mini-tts is the token-billed model with streaming output and steerable voice instructions; In October 2026 Voice Arena measured it at an 811.5ms median time-to-first-audio. The standout is gpt-realtime-2.1, released on 6 July 2026 with configurable reasoning effort and at least 25% lower p95 latency than its predecessor, which handles speech-to-speech directly — you send audio and reasoning and speech come back together, removing the separate text-to-speech step (OpenAI).

Why it wins: The simplest path if you already build on OpenAI, and the cleanest on-ramp to speech-to-speech agents where reasoning and voice live in one model.

Limitations: The Realtime models lock you to OpenAI’s stack, and the dedicated agent APIs measure far faster to first audio than gpt-4o-mini-tts. Voice cloning is not offered on these endpoints.

Best for: Products already in the OpenAI ecosystem, and teams building speech-to-speech agents rather than a best-of-breed pipeline.

9. Hume — best for emotion-critical delivery

Type: Expressive Flagship: Octave 2 (text-to-speech) and EVI (empathic voice interface) Pricing: Usage-based API, with plans from $3 to $500 per month; Octave 2 cut costs roughly 50% versus the prior generation Key features: Explicit emotional control via instructions, context-aware delivery, 11 languages, multi-speaker conversation, real-time empathic voice interface

Hume splits its offering into Octave, a context-aware text-to-speech model, and EVI, a real-time speech-to-speech model that detects emotional cues and responds in kind. Its differentiator is explicit emotional control: you direct delivery with instructions rather than relying on voice choice alone. The trade-off is that this surface is useful when emotion is the point, but for most workloads ElevenLabs or Google deliver enough expressiveness through voice selection and direction prompts alone.

Why it wins: The most direct programmatic control over emotional delivery, in both text-to-speech and speech-to-speech, for applications where how something is said matters as much as what is said.

Limitations: On raw naturalness it trails the leaderboard leaders, Octave 2 covers 11 languages, and the emotional-control surface adds complexity you do not need for neutral narration or standard agents.

Best for: Wellness and therapy apps, character voices and empathetic agents where emotional delivery is the load-bearing feature.

10. Microsoft Azure AI Speech — best for enterprise compliance and language breadth

Type: High-volume Flagship: Neural and Neural HD voices, with Dragon HD Omni in preview Pricing: $15–16 per million characters (neural); $22 per million (Neural HD, cut from $30 in March 2026); $24 per million for custom voice; commitment tiers cut the effective rate to about $7.50 per million; $200 credit plus 500K characters per month free Key features: 500+ voices across 140+ languages, custom neural voice, HIPAA BAA, SOC certification and container deployment

Microsoft Azure AI Speech offers the widest language coverage and the strongest compliance story of the mainstream APIs. Its newest voice type, Dragon HD Omni, is in preview through Microsoft Foundry with 700+ voices, style control and automatic style prediction; in October 2026 Voice Arena measured it at a 485ms median time-to-first-audio. Unlike Amazon Polly, Azure supports custom voice cloning at $24 per million characters. For organisations standardised on Microsoft — Entra ID, Azure governance, in-region data — it puts high-quality synthesis behind identity, logging and data-residency controls you already run, with container deployment available for on-premise needs.

Why it wins: The broadest language coverage and the deepest compliance and enterprise-governance posture, inside the Microsoft cloud.

Limitations: Streaming latency is moderate rather than class-leading, Dragon HD Omni is preview-only in a few regions, and the best value assumes you are already an Azure organisation with the volume to reach a commitment tier.

Best for: Regulated industries, healthcare and multilingual telephony where compliance and language breadth outweigh peak latency.

11. Amazon Polly — best for high-volume AWS-native synthesis

Type: High-volume Flagship: Neural and Generative voices Pricing: $4 per million characters (standard), $16 (neural), $30 (generative), $100 (long-form); 5M standard or 1M neural characters free for 12 months Key features: Mature SDKs, 99.9% uptime SLA, deep AWS integration, LLM-based Generative voices

Amazon Polly is the high-volume workhorse for AWS-native applications. Its Generative voices use LLM-based synthesis at $30 per million characters, though they are English-only with no streaming, while standard neural voices at $16 per million cover the everyday case cheaply. Polly’s advantages are operational: mature SDKs, a 99.9% uptime SLA and cost predictability inside AWS. It has no voice cloning, which rules it out for branded or custom-voice work.

Why it wins: The most predictable, AWS-native path for large-scale synthesis, with a strong SLA and the lowest entry price for standard voices.

Limitations: No voice cloning, English-only Generative voices, and quality that trails the leaderboard leaders. It is an infrastructure choice, not a quality one.

Best for: IVR, accessibility features and any high-volume application where AWS integration and cost predictability beat cutting-edge quality.

Feature comparison: the full matrix

Data-retention and training policies change; confirm the current terms in each vendor’s documentation before handling regulated or personal data. Latency figures are vendor-reported time-to-first-audio or time-to-first-byte unless marked Voice Arena (measured in October 2026), and vary by region and payload — measure on your own traffic before committing.

Pricing comparison: what you’ll actually pay

Per million characters (list, USD)

The Google per-character equivalents are derived from Google’s own per-10-second audio figures at roughly 54,000 characters per hour of speech.

Cost by audio duration

One hour of speech is roughly 9,000 words, about 54,000 characters. The table below estimates character- or token-billed API cost at that rate on list prices; subscription and credit plans bundle a fixed allocation per plan instead.

Google’s Gemini 3.8 figures double from 1 January 2027, and ElevenLabs v4 costs about 72% less than shown until 12 October 2026.

Cost strategy: route by job, not loyalty. Use a premium model such as Eleven v4 only where quality drives revenue, send the live conversational turn to a real-time API priced for volume, push high-volume, latency-tolerant synthesis (notifications, batch narration) to a cheap cloud model like Gemini Flash-Lite TTS or Amazon Polly, and self-host an open-weight model for privacy-sensitive or offline work. The blended rate, not the headline number, is what moves your bill at scale — and on token-billed models, estimate from audio duration rather than text length.

Latency: streaming, time-to-first-audio and voice agents

For anything a human waits on in real time, the number that matters is time-to-first-audio: how long after you send text the first sound comes back. Below roughly 300ms, turn-taking in a conversation feels natural; above it, the agent feels laggy.

Vendor figures and independent measurements now tell different stories, so the table shows both. Vendor numbers are usually server-side or model-only; In October 2026 Voice Arena reported a median time-to-first-audio it measured itself across 14 models.

Use streaming, low-latency APIs for voice assistants, phone agents, live translation and gaming dialogue. Use batch, file-return synthesis for notifications, e-learning, audiobooks and marketing audio, where a few seconds of generation time is irrelevant and cost matters more. Independent voice-AI evaluators caution that vendor latency and quality numbers are measured under favourable conditions, so test on your own audio and traffic before paying a premium (Coval).

Use-case recommendations

For real-time voice agents

Winner: Cartesia Sonic 3.6 or Inworld Realtime TTS-2

Both had top-five quality on the Artificial Analysis leaderboard on 29 September 2026, with streaming built for agents: Cartesia for the “bring your own speech-to-text and language model” pattern with pinned model snapshots, Inworld for natural-language steering, 200+ languages and the lower independently measured latency. Alternatives: SpeechifyAI Simba 3.2 for English-only agents, where Voice Arena measured it fastest to first audio in October 2026, at $6–10 per million characters; Deepgram Flux TTS if you already transcribe with Deepgram; Rime when conversational phone prosody or on-premise deployment is the priority.

For the best voice quality

Winner: ElevenLabs API with Eleven v4

First on the Artificial Analysis leaderboard on 29 September 2026, by 43 Elo points, with 90+ languages and the deepest cloning, for produced audio where quality drives revenue. Use v4 Turbo for the real-time tier of the same account. Alternative: Google Gemini 3.8 Flash TTS, third on that leaderboard on 29 September, for a fraction of the cost when budget outranks the last few points.

For the cheapest synthesis at scale

Winner: Google Gemini 3.8 Flash-Lite TTS ($6 per million audio tokens until 31 December)

Roughly $0.09 for a 10-minute narration at top-ten quality across 100+ languages, rising to about $0.18 when the price doubles on 1 January 2027. Alternatives: SpeechifyAI for English, at $10 per million characters on Starter and $6 on the $499/month Scale plan, which undercuts Flash-Lite at volume; and Amazon Polly standard voices at $4 per million characters for AWS-native, quality-tolerant volume.

For emotion-critical delivery

Winner: Hume Octave 2

Explicit, instruction-driven emotional control for wellness apps, characters and empathetic agents where delivery matters as much as content. Alternative: ElevenLabs Eleven v4, whose natural-language direction and audio tags now cover much of the same ground at higher overall quality.

For enterprise compliance and language breadth

Winner: Microsoft Azure AI Speech

140+ languages, custom neural voice, HIPAA BAA and SOC certification inside the Microsoft cloud — the safe default for regulated, multilingual telephony.

For speech-to-speech agents

Winner: OpenAI gpt-realtime-2.1

When you want reasoning and voice in one model rather than a separate TTS step, gpt-realtime-2.1 is the simplest path, and the gpt-realtime-2.1-mini variant cuts the audio-token price to $10/$20 per million. For a fully bundled, no-code phone agent, SpaceXSI Grok Voice ships one, with Grok Voice 3.0 at $0.08 a minute (SpaceXSI) and, in October 2026, the highest task-success rate in the Artificial Analysis Speech Agent Arena; see best AI voice agents.

For self-hosting and data control

Winner: Kokoro-82M or Chatterbox

Both run locally with no per-character cost. Kokoro-82M (Apache 2.0, 82M parameters) is the best default for speed and edge deployment and runs faster than real time on modest hardware; Chatterbox (MIT) clones from a 10-second clip but is English-only and watermarks output. For permissive cloning, Qwen3-TTS (Apache 2.0) clones from 3-second samples.

How to choose a text-to-speech API

Nine dimensions decide the fit. Rank them for your build before you compare vendors.

Latency — for anything real-time, time-to-first-audio is the load-bearing number, and in October 2026 Voice Arena’s independent measurements ran three or four times the vendor figures. Streaming protocol — confirm the API supports chunked or WebSocket streaming, not just file return, if a human is waiting on the audio. Price model — character billing ($4–100 per million) is easy to forecast, while token billing on the newest Google and OpenAI models needs estimating from audio duration; watch for scheduled price rises such as Google’s in January 2027. Voice library and cloning — do you need a stock catalogue, a custom brand voice, or instant cloning, and does the vendor support it. Languages — coverage ranges from English-only (SpeechifyAI Simba 3.2, Deepgram Flux TTS) to 140+ (Azure) and a vendor-stated 200+ (Inworld); match it to your markets. Control — SSML, emotion tags and natural-language steering vary widely; Hume, Inworld and ElevenLabs v4 lead on explicit control. Concurrency and rate limits — high-volume and real-time workloads hit different limits, so check the concurrent-stream caps before launch. Data and compliance — confirm training, retention, HIPAA, SOC and residency terms in writing for regulated data. Self-host option — if audio cannot leave your network, an open-weight model (Kokoro, Chatterbox) or an on-premise vendor (Rime, Deepgram, Azure containers) is the only path.

Our picks are ranked by suitability for a new voice build, weighting quality, latency, streaming, price and language coverage. We anchor every recommendation in vendor-documented figures and independent leaderboards — the Artificial Analysis Text-to-Speech leaderboard for quality and Voice Arena for measured latency — and we treat vendor latency and naturalness claims as a ceiling to verify against your own traffic, not a floor.

Building a voice agent: where the TTS API fits

A custom voice agent is a three-stage pipeline, and text-to-speech is the last stage. Speech-to-text transcribes what the user said, a language model decides what to say back, and a text-to-speech API speaks the reply. Each stage is a separate API decision, and the latency of all three stacks up into the total response time the user feels.

For the speech-to-text stage, see best AI for transcription, which compares Deepgram, AssemblyAI, Mistral Voxtral and the other speech-to-text APIs on word error rate and streaming latency. For the language-model stage, see best LLM APIs, which ranks OpenAI, Anthropic, Google and the aggregators on price, speed and model quality. To orchestrate the three stages, an agent framework helps — see best AI agent frameworks.

The alternative is to collapse all three stages into one speech-to-speech model such as OpenAI gpt-realtime-2.1 or SpaceXSI Grok Voice, which removes the separate text-to-speech step entirely. The three-stage pipeline wins when you want to control the exact voice, mix best-of-breed components or cut cost on high-volume synthesis; the single-model approach wins when you want the simplest build and lowest integration overhead.

Related guides

For choosing a voice tool with a web interface rather than an API — produced voice-overs, a voice library, a reader — see best AI voice generator. For turning speech into text, see best AI for transcription. For voice cloning specifically and the legal landscape around it, see best AI voice clone. For the wider developer stack, see best LLM APIs, best AI agent frameworks and the best AI apps overview.

Frequently asked questions

What is the best text-to-speech API in 2026?

There is no single best text-to-speech API — the right choice depends on the job. For the best voice quality, the ElevenLabs API with Eleven v4, which was first on the Artificial Analysis leaderboard on 29 September 2026. For real-time voice agents, Cartesia Sonic 3.6 and Inworld Realtime TTS-2 paired top-five quality on that date with low latency, and SpeechifyAI Simba 3.2 was the fastest independently measured option for English in October 2026. For cheap top-tier synthesis across many languages, Google Gemini 3.8 Flash-Lite TTS costs $6 per million audio tokens until the end of 2026. Match the API layer — real-time, expressive or high-volume — to your build before comparing vendors.

What is the cheapest text-to-speech API?

Among high-quality options, Google Gemini 3.8 Flash-Lite TTS and SpeechifyAI Simba are the cheapest. Flash-Lite costs $6 per million audio tokens — roughly $0.09 for a 10-minute narration — until 31 December 2026, after which it doubles to $12. SpeechifyAI costs $10 per million characters on Starter, about the same per hour, falling to $6 on its Scale plan, with 500,000 free characters a month, though its top model is English-only. Amazon Polly standard voices are $4 per million characters but lower quality, and Inworld reaches $5 per million characters or less at enterprise scale. For zero per-character cost, the open-weight Kokoro-82M model runs locally on modest hardware.

Which text-to-speech API has the lowest latency?

It depends on whose measurement you use. On vendor figures, Inworld Realtime TTS-2 Flash claims 25ms time-to-first-byte (server-side, P99), Deepgram Flux TTS as low as 80ms and Cartesia Sonic 3.6 sub-90ms model latency. On independent measurement in October 2026, Voice Arena ranked SpeechifyAI Simba 3.2 fastest at a 123ms median time-to-first-audio, followed by Inworld Realtime TTS-2 at 168.5ms, with Cartesia Sonic 3.6 at 340.5ms. For natural conversational turn-taking, aim for a time-to-first-audio under 300ms measured on your own traffic.

What is the best text-to-speech API for voice agents?

For text-to-speech-only voice agents where you supply your own speech-to-text and language model, Cartesia Sonic 3.6 and Inworld Realtime TTS-2 combine top-five quality with streaming built for agents, and SpeechifyAI Simba 3.2 is the lower-cost English-only option with the fastest measured latency. Deepgram Flux TTS suits teams on the Deepgram stack, and Rime specialises in conversational phone-agent prosody with an on-premise option. If you would rather use one model for the whole pipeline, OpenAI gpt-realtime-2.1 handles speech-to-speech directly.

What is the difference between a text-to-speech API and a voice generator tool?

A text-to-speech API is an endpoint you call from code to synthesise speech inside a product, billed per character or token and measured on latency, streaming and control. A voice generator tool is an application with a web interface, a voice library and an editing workflow, aimed at producing voice-overs, audiobooks or narration without writing code. Developers building agents, captions or in-app speech want an API; creators producing content want a tool. This guide ranks the APIs; our best AI voice generator guide ranks the tools.

Can I run a text-to-speech API locally for privacy?

Yes, and open-weight models are now genuinely competitive. Kokoro-82M (Apache 2.0, 82M parameters) runs faster than real time on modest hardware or CPU with no per-character cost, Chatterbox (MIT) clones a voice from a 10-second clip, and Qwen3-TTS (Apache 2.0) offers permissive cloning from 3-second samples. Self-hosting keeps all audio on your own machine, which removes the data-residency and retention concerns of cloud APIs. The trade-off is that you run and scale the infrastructure yourself.

Do text-to-speech APIs support streaming?

Every agent-grade text-to-speech API in 2026 supports streaming — chunked HTTP or WebSocket delivery so audio starts playing before the full sentence is synthesised. Cartesia, Inworld, SpeechifyAI, Deepgram, Rime and ElevenLabs v4 Turbo all stream, which is what makes sub-200ms perceived latency possible. Older, file-return endpoints that synthesise the whole clip before returning it still exist for batch narration but are not suitable for live conversation, where the first audio byte needs to arrive within a few hundred milliseconds.

Is it legal to clone a voice with a text-to-speech API?

Synthesising speech from text you have written is legal. Voice cloning is where the law applies: cloning your own voice or a voice you have documented consent for is legal, while cloning someone without consent can violate laws such as the Tennessee ELVIS Act, New York’s Digital Replicas Law and the FCC’s robocall order. Vendors are building consent checks into the product — ElevenLabs v4 requires verified consent from the voice’s owner, and Google’s Gemini 3.8 TTS requires a matching consent recording and withholds replication in several jurisdictions. As of 2026, active class actions are testing how voice-AI training and cloning are regulated. If cloning is central to your product, read our best AI voice clone guide for the detail, and treat a consent checkbox as insufficient legal protection.

How much does a text-to-speech API cost for a real product?

Character-billed APIs run $4 to $100 per million characters, and one hour of speech is about 54,000 characters. At that rate, 100 hours a month costs roughly $54 on Google Gemini 3.8 Flash-Lite TTS (until its January 2027 price rise) or SpeechifyAI Starter, $86 on Amazon Polly Neural, $162 on Deepgram Aura-2 or about $432 on ElevenLabs Eleven v4 at list price. Real-time agent traffic is usually short per turn but high in volume, so model your cost from expected minutes of audio, not documents. Routing the live turn to a premium model and batch synthesis to a cheap one keeps the blended rate down.