Top 10 Text-to-Speech API Providers of 2026

You test five text-to-speech APIs in a demo and they all sound fine. Then production starts and the differences show up: one bills per character, one in credits, one in audio tokens, and the plan you picked caps how many requests can run at once.

This guide compares the 10 best text-to-speech API providers of 2026 with current list prices, free tiers and concurrency limits, so you can shortlist on what speech will actually cost you at your volume.

What is a text-to-speech API?

A text-to-speech (TTS) API turns written text into spoken audio over an HTTP or streaming interface. You send characters, the service returns audio, and you pay for what you synthesize. Modern TTS APIs cover a wide range: simple narration voices, emotionally expressive models, voice cloning, and low-latency streaming voices built for real-time voice agents. The provider you choose sets your voice quality, your latency budget and your monthly bill.

How we selected these text-to-speech API providers

We shortlisted providers that developers actually put into production in 2026: the major cloud speech services, the specialist voice AI vendors, and the API-first voice agent suppliers. Every provider below has a public API and public pricing we could verify on its official pricing page in November 2026. We compared billing units (characters, credits or tokens), list price per million characters, free tiers, concurrency rules and model lifespan, because a cheap model scheduled for shutdown is not cheap.

Quick comparison of the best text-to-speech APIs

List prices are normalized to 1 million characters for comparison. Deepgram and Murf quote native rates per 1,000 characters; Rime bills per minute of audio, normalized here at roughly 750 to 1,000 characters per minute of speech; ElevenLabs API pricing is per 1,000 characters, while its subscription plans meter usage in monthly credits (the free plan includes 10K credits). Sources for every rate are linked under Sources below.

List prices as published on official pricing pages, checked 12 November 2026.

1. ElevenLabs

ElevenLabs is the quality benchmark most teams compare against. Its flagship models (Eleven v4, v3 and Multilingual v2) focus on expressive, emotionally rich speech, while the Turbo and Flash models trade some expressiveness for speed, with v4 Turbo rated at roughly 100ms latency. The API supports professional voice cloning, 90+ languages on the multilingual models, dubbing and speech-to-speech conversion. Plans bundle monthly credits that all products share, so the same balance can fund TTS, dubbing or speech-to-text.

Best for: teams that put voice quality and cloning first, and products where the voice is part of the brand.

Pricing: API usage is billed per 1,000 characters: $0.08 ($80 per 1M) for v4, v3 and Multilingual v2, and $0.04 ($40 per 1M) for v4 Turbo, v3 Conversational and Flash/Turbo. Subscription plans run from Starter at $6 per month (30K credits) through Creator $22, Pro $99 and Scale $299 to Business $990, with 10K credits free each month. Concurrency is capped per plan, from 2 simultaneous requests on Free to 15 on Scale and Business. View ElevenLabs pricing.

2. OpenAI

OpenAI's TTS models are the obvious pick if your stack already runs on OpenAI. The catch is timing: OpenAI's deprecations page lists tts-1, tts-1-hd and gpt-4o-mini-tts for shutdown on January 6, 2027, with the token-billed realtime model gpt-realtime-2.1-mini as the recommended replacement. Starting a new build on the current TTS models means planning a migration within months, and the replacement bills audio tokens instead of characters, so your cost model changes with the model name.

Best for: teams already invested in the OpenAI platform who are comfortable migrating to realtime models.

Pricing: tts-1 costs $15 per 1M characters and tts-1-hd $30 per 1M characters. gpt-4o-mini-tts bills $0.60 per 1M text input tokens plus $12 per 1M audio output tokens. View OpenAI pricing and the OpenAI deprecations page.

3. Google Cloud Text-to-Speech

Google's Cloud Text-to-Speech is the predictable cloud option: plain per-character billing, a genuinely useful free tier, and a wide voice ladder from cheap Standard voices to Chirp 3: HD and Studio voices. Google also offers Gemini TTS models billed per token, in preview. The fine print worth knowing: Google counts spaces, newlines and most SSML tags as billable characters.

Best for: teams that want cloud-grade reliability, a large free tier for development, and simple forecasting.

Pricing: Standard and WaveNet voices cost $4 per 1M characters with the first 4M free each month. Neural2 voices are $16 per 1M and Chirp 3: HD voices $30 per 1M, each with 1M free characters monthly. Studio voices run $160 per 1M and instant custom voice $60 per 1M. Gemini 3.8 Flash TTS, in preview, bills $0.50 per 1M text input tokens plus $9 per 1M audio output tokens through December 31, 2026, with the listed price doubling on January 1, 2027. View Google Cloud TTS pricing.

4. Amazon Polly

Amazon Polly is the AWS-native choice, with four voice tiers that map to different budgets: Standard for cheap utility speech, Neural for natural voices, Generative for the most expressive output, and Long-Form voices built for narration like articles and audiobooks. If your app already lives on AWS, Polly keeps billing, IAM and regions in one place.

Best for: AWS-native products, and long-form narration where the Long-Form voices fit.

Pricing: Standard voices cost $4 per 1M characters, Neural $16, Generative $30 and Long-Form $100. The free tier, for the first 12 months, includes 5M Standard characters per month, 1M Neural, 500K Long-Form and 100K Generative. View Amazon Polly pricing.

5. Microsoft Azure AI Speech

Azure's Speech service covers text to speech alongside speech to text, translation and avatars, which suits enterprises standardizing on one Microsoft vendor. Text-to-speech billing is per character with tiered volume commitments, and the catalog includes prebuilt neural voices, Neural HD voices, custom professional voices and Personal Voice. As with other cloud consoles, exact per-million rates depend on region and commitment, so confirm them in the Azure pricing calculator for your deployment.

Best for: enterprises on the Microsoft stack that want one vendor for speech in both directions.

Pricing: Neural text to speech bills per 1M characters, with 0.5M characters free per month and lower effective rates through monthly commitment tiers. View Azure Speech pricing.

6. Cartesia

Cartesia is built for real-time voice agents. Its Sonic model family targets low time-to-first-audio, and the pricing model is simple once you learn the unit: one credit equals one character, and plans are monthly blocks of credits. Cartesia puts one minute of Sonic speech at roughly 750 to 800 credits, which makes per-minute math easy. Instant voice cloning is included even on the $5 plan.

Best for: developers building latency-sensitive voice agents who want plan-based credits instead of cloud billing.

Pricing: Free includes 20K credits per month. Pro is $5 for 100K credits, Startup $49 for 1.25M and Scale $299 for 8M. Overage runs $65, $45 and $38 per 1M credits respectively. TTS concurrency is fixed per plan: 2 requests on Free, 3 on Pro, 5 on Startup and 15 on Scale. View Cartesia pricing.

7. Deepgram (Aura and Flux)

Deepgram sells Aura, its streaming TTS family, alongside the speech-to-text models many voice-agent stacks already use, so one vendor can cover both directions of the call. Aura-2 is the current generation, Aura-1 the cheaper legacy option, and Flux TTS the premium streaming model. Billing is per 1,000 characters on prepaid credits, with lower rates on the Growth plan, which starts at $4,000 a year in prepaid credits.

Best for: voice-agent teams that want STT and TTS from one provider with one bill.

Pricing: Pay as you go: Aura-2 at $0.030 per 1K characters ($30 per 1M), Aura-1 at $0.015 ($15 per 1M) and Flux TTS at $0.045 ($45 per 1M). Growth drops Aura-2 to $0.027 and Aura-1 to $0.0135 per 1K. View Deepgram pricing.

8. AssemblyAI

AssemblyAI, long known for speech-to-text, now ships its own TTS model, AssemblyAI-TTS-1, aimed squarely at voice agents. The pitch is accurate read-out of the things agents actually say, like phone numbers, prices, dates and emails: on AssemblyAI's published evaluations, TTS-1 scored 93.7% entity accuracy, fractionally ahead of ElevenLabs v4 at 93.6%. Cloned voices and EU processing carry no surcharge, and word-level timestamps are included. As with any vendor-run benchmark, treat the numbers as a starting hypothesis and test on your own scripts.

Best for: voice agents where misreading a number costs a repeat turn or a lost caller.

Pricing: AssemblyAI-TTS-1 is $15 per 1M input characters on pay as you go. Capacity is limited by new sessions per minute rather than open sessions: 5 per minute on the free plan and 200 on paid plans. View AssemblyAI pricing.

9. Murf AI

Murf runs two API models at two price points: Falcon 2.1, optimized for conversational AI and real-time voice agents with time-to-first-audio under 100ms, and Gen2, its studio-quality model for rich voiceovers. Beyond TTS, the same account covers speech-to-text, voice changing and translation, and Murf also sells a creative Studio product for content teams. Enterprise plans add on-premise deployment and professional voice cloning.

Best for: teams that want one vendor for fast conversational TTS plus studio voiceover work.

Pricing: Pay as you go: Falcon 2.1 at $0.01 per 1K characters ($10 per 1M) and Gen2 at $0.03 per 1K ($30 per 1M). Speech-to-text bills $0.12 per hour and Voice Changer $0.10 per minute. Enterprise pricing is custom. View Murf pricing.

10. Rime

Rime builds speech models for real-time voice agents, especially phone and customer-service calls. Its current lineup is Mist v3 for low-latency text to speech and Coda v2, the newer flagship that now serves the retired Arcana model's traffic, plus Prism for speech-to-speech conversations and Rime STT for speech to text. Billing is per minute of audio, with no seat plan or studio subscription to buy first.

Best for: teams building phone and voice-agent products that want simple per-minute pricing.

Pricing: On the pay-as-you-go Flex plan, Mist v3 costs $0.030 per minute of audio and Coda v2 $0.050 per minute, with the Prism speech-to-speech model at $0.080 per minute. The Scale plan at $100 per month cuts every rate by 20% (Mist v3 $0.024, Coda v2 $0.040) and its monthly credit rolls over, the Free plan includes $50 of usage every month, and Enterprise is custom. View Rime pricing and the Arcana to Coda migration note.

How to choose the right text-to-speech API

Pick the billing model first, then the vendor. Four steps:

1. Measure your characters per call or per asset. Pull a week of real agent replies or scripts and count. Every price above multiplies off this number. A minute of agent speech runs roughly 750 to 1,000 characters depending on speaking rate.
2. Find your peak concurrency. Your busiest minute, not your average, sets the plan. Cartesia and ElevenLabs publish hard caps per plan; the clouds govern throughput with account quotas. Test at real peak before launch.
3. Check the calendar. OpenAI's TTS models shut down on January 6, 2027, and Google's Gemini TTS preview price doubles on January 1, 2027. A cheap model with a dated change is a migration project, not a saving.
4. Price the whole call, not the TTS line. For voice agents, TTS sits next to streaming speech-to-text and the LLM. A low TTS rate in a stack that mishears callers is not a saving.

If you want a shortcut, this table maps common situations to the providers above that fit them best. It is a starting shortlist, not a ranking.

What to check before comparing TTS pricing

• Billing unit: per character (easy to forecast), credits inside plans (capacity tied to spend), or audio tokens (cost scales with speech length, not text length).
• What counts as a character: some providers bill spaces, newlines and SSML tags.
• Rollover and expiry: ElevenLabs rolls unused credits for up to two months; Cartesia rolls up to 2x the monthly allowance; cloud free tiers reset monthly.
• Concurrency and queueing: a plan capped at 5 simultaneous requests turns your growth into caller dead air.
• Model lifespan: preview pricing, promotions and shutdown dates all change your 2027 cost.
• Extras that cost extra elsewhere: voice cloning, EU processing and word-level timestamps are bundled by some providers and billed by others.

If you are comparing voice-agent stacks end to end, see our roundup of the best AI voice agent platforms in 2026, our guide to safe and realistic AI voice cloning, and our explainer on AI tool pricing models for the credit-and-token math in more depth.

FAQs

Which text-to-speech API is the cheapest in 2026?

On list price, Google Cloud Standard voices and Amazon Polly Standard voices are the cheapest at $4 per 1M characters, and both carry meaningful free tiers. For natural-sounding neural voices, Murf's Falcon 2.1 lists lowest at $10 per 1M, ahead of AssemblyAI-TTS-1, OpenAI's tts-1 and Deepgram Aura-1 at $15, though OpenAI's tts-1 retires on January 6, 2027. Compare at your real volume in characters, because free tiers and plan bundles move the answer.

What is the difference between per-character, credit and token pricing?

Per-character pricing bills the text you send, which is the easiest to forecast. Credit plans sell you a monthly block where one credit is roughly one character, tying capacity to your plan. Token pricing, used by OpenAI's and Google's realtime models, bills text input tokens plus audio output tokens, so cost scales with how long the speech runs, not how much text you sent.

Is OpenAI shutting down its TTS API?

Yes. OpenAI's deprecations page lists tts-1, tts-1-hd and gpt-4o-mini-tts for shutdown on January 6, 2027. The recommended replacement is gpt-realtime-2.1-mini, a realtime model billed per token, so migrating changes both your model and your meter.

Which TTS API has the best free tier?

Google Cloud offers the largest standing free tier: 4M Standard or WaveNet characters and 1M Neural2 or Chirp 3: HD characters free every month. Amazon Polly's free tier covers 5M Standard characters per month for your first 12 months. Cartesia and ElevenLabs include 20K and 10K monthly credits respectively, and Azure includes 0.5M Neural characters per month.

What matters more than price when choosing a TTS API?

Three things: time to first audio (what the listener perceives as responsiveness), entity accuracy (reading phone numbers, prices and dates correctly), and concurrency (how many requests run at once before callers hear a queue). A cheap minute of speech that reads an order number wrong costs you a repeat turn and sometimes the caller.

Final verdict

There is no single best text-to-speech API in 2026, but there is a best fit per workload. For brand-defining voice quality, start with ElevenLabs. For low-latency voice agents, benchmark Cartesia, Deepgram Aura, AssemblyAI and Murf at your real peak. For cheap, predictable cloud billing, Google and AWS are hard to beat, and their free tiers cover most development. Whatever you choose, model the cost at your real character volume and your busiest minute before you commit.

When you are ready to compare specific tools, ToolJunction is a discovery platform for AI startups, tools and agents, with pricing breakdowns and comparisons across categories.

Sources

• ElevenLabs pricing and ElevenLabs API pricing
• OpenAI API pricing and OpenAI model deprecations
• Google Cloud Text-to-Speech pricing
• Amazon Polly pricing
• Microsoft Azure AI Speech pricing
• Cartesia pricing
• Deepgram pricing
• AssemblyAI pricing and the AssemblyAI TTS pricing analysis
• Murf AI pricing
• Rime pricing and the Rime Arcana migration note

Prices checked 12 November 2026 on the official pages linked above. Prices change; verify before quoting.