Back to Blog
Developers

Amazon Polly Alternatives in 2026: Limits and Options

Amazon Polly alternatives for 2026: what Polly does well, where its engines, limits and regions fall short, and how Google, Azure and OpenAI compare.

10 min read
By OndaLabs Team

Amazon Polly alternatives are not all alike, so the best one depends on what Polly is failing to do for you. Polly's Standard engine lists at $4 per 1 million characters, but its expressive engines have narrower regions, shorter voice lists and tighter throughput limits. As of October 2026, Google, Azure and OpenAI publish per-character or per-token prices you can compare directly.

This guide reviews what Polly does well, where it runs into limits, how the alternatives price their voices and which job fits which service. Every price was checked on the vendor pages on 2026-10-17.

What Amazon Polly does well

Polly offers four engines: Standard, Neural, Long-Form and Generative. It also supports speech marks, which return sentence, word and sound timing alongside the audio, plus custom lexicons and SSML. Speech marks are a meaningful advantage when your product has to highlight text as it is spoken.

AWS says Polly is "certified for use with regulated workloads for HIPAA ... and PCI DSS." It also says you can cache and replay generated speech at no additional cost. AWS's Polly documentation lists these points, and they matter if you store generated audio for a product.

Polly's features page describes "dozens of lifelike voices" and a Newscaster style for a small set of voices. Brand Voice is a custom engagement that builds a Neural voice for one organization's exclusive use. Polly's docs do not describe self-serve voice cloning, so do not assume it exists.

Amazon Polly pricing

The Polly pricing page, fetched on 2026-10-17, lists pay-as-you-go rates per 1 million characters:

| Engine | List price per 1M characters | Free tier | Speech marks | Notes | |---|---|---|---|---| | Standard | $4.00 | 5 million characters per month | Yes | Many languages and variants | | Neural | $16.00 | 1 million per month for the first 12 months | Yes | 36 languages and variants, 19 regions | | Long-Form | $100.00 | 500,000 per month for the first 12 months | Yes | Six voices, US East (N. Virginia) only | | Generative | $30.00 | 100,000 per month for the first 12 months | No | 43 voices, 10 regions, speech requests only |

The $4 headline price is for Standard. Neural and Generative cost four and seven and a half times as much per character, so do not assume a Neural voice costs $4.

AWS converts 1 million characters to about 23 hours and 8 minutes of speech. Its example, "A Christmas Carol" at about 165,000 characters (roughly 3 hours 50 minutes of audio), costs $0.66 on Standard, $2.64 on Neural, $16.50 on Long-Form and $4.95 on Generative. From those examples, Neural works out to about $0.69 per hour of audio and Generative to about $1.30 per hour, both derived.

GovCloud (US) pricing is higher: Standard $4.80 and Neural $19.20 per 1 million characters. New AWS customers from 2025-07-15 can receive up to $200 in Free Tier credits.

Where Polly runs into limits

Polly's published quotas are where most migrations start. The limits page and the generative voices page list these constraints:

  • Synchronous requests: up to 3,000 billed characters per request, or 6,000 characters total. Output for a stream is limited to 10 minutes.
  • Asynchronous tasks: up to 100,000 billed characters per task, or 200,000 total.
  • Throughput: the default is 80 transactions per second for Standard, but only 8 for Neural, Long-Form and Generative. Asynchronous tasks have lower limits for generative and long-form voices.
  • Generative regions: the engine runs in 10 listed regions, and London, Canada Central and Zurich offer only a subset of 22 named voices.
  • Generative features: speech marks are not available, and Newscaster style is not supported.
  • Long-Form: only US East (N. Virginia) hosts the engine, with six voices (four en-US and two es-ES).
  • Hallucination safeguards: AWS says the generative engine has a safety stop that "reduces but does not eliminate the risk" of hallucinated output. AWS also says model updates "could result in slight variations" in how voices sound over time.

SSML tags are not billed on Polly. That is an important detail when you compare prices, because Azure bills SSML markup and Google counts SSML tags except the mark element.

Amazon Polly alternatives compared

This table uses published list prices, converted to a per-1-million-character basis where the vendor bills per 1,000 characters. Conversions are derived and labelled.

| Service | List price per 1M characters | Billing basis | Free tier | |---|---|---|---| | Amazon Polly Standard | $4 | Characters | 5M per month | | Google Standard and WaveNet | $4 | Characters (SSML tags counted, except mark) | 4M per month | | Amazon Polly Neural | $16 | Characters | 1M per month, 12 months | | Google Neural2 | $16 | Characters | 1M per month | | Azure Neural | $15 | Characters (SSML markup counted) | 0.5M per month (F0) | | OpenAI tts-1 | $15 | Characters | None stated | | OpenAI tts-1-hd | $30 | Characters | None stated | | Azure Neural HD | $22 | Characters | None stated | | Google Chirp 3 HD | $30 | Characters | 1M per month | | Amazon Polly Generative | $30 | Characters | 100K per month, 12 months | | Deepgram Aura-2 | $30 (derived from $0.030 per 1,000) | Per 1,000 characters | $200 credit | | ElevenLabs v4 Turbo | $40 (derived from $0.04 per 1,000) | Per 1,000 characters | Free plan on elevenlabs.io | | ElevenLabs v4 | $80 (derived from $0.08 per 1,000) | Per 1,000 characters | Free plan on elevenlabs.io |

Sources: Google Cloud text to speech pricing, OpenAI API pricing, ElevenLabs API pricing and Deepgram's pricing page, all checked on 2026-10-17. Azure figures come from the Azure Retail Prices API for East US.

Three cautions apply to the table:

  • OpenAI's gpt-4o-mini-tts is billed per token, so it is left out of the per-character comparison.
  • Google's Gemini TTS models are also billed per token. See our Gemini text to speech guide for those rates.
  • Google Chirp 3 HD offers 30 voices and 53 locales, and streaming is supported. SSML is not supported for streaming requests.

Quality and speed claims

Artificial Analysis publishes a TTS leaderboard that uses user votes. On its page, Polly Standard is named the fastest by throughput at 1,339.2 characters per second through Amazon Bedrock, and ElevenLabs v4 Turbo is named the highest quality, with a Quality Elo of 1330. The page does not give a date beyond 2026, and these are the results of one arena, not tests run for this guide.

Polly does not publish official latency numbers for its engines. Treat any single throughput or quality number as one data point, and run your own blind listening test on the voices your users will hear.

Which alternative fits which job

  • Speech marks and timed highlighting: stay on Polly Standard, Neural or Long-Form, since Generative does not return speech marks.
  • Lowest list price per character: Google Standard and WaveNet, Polly Standard and Azure Neural are all in the $4 to $15 range, depending on the vendor and tier.
  • Instruction-driven delivery: OpenAI's gpt-4o-mini-tts takes an instructions parameter, and Google's Gemini TTS takes a style field and inline tags.
  • Real-time voice agents: ElevenLabs describes v4 Turbo as its low-latency variant for real-time use. Polly Generative offers a bidirectional streaming API in eight regions, at 8 transactions per second.
  • Finished voiceover and dialogue: a voice studio may fit better than an API. OndaLabs is one option for that work.

Switching from Polly

Plan the migration around the differences that break code first:

  • Billing of SSML: Polly does not bill SSML tags, but Azure bills SSML markup and Google counts most SSML tags. The same script can cost more on another vendor.
  • Speech marks: if you depend on them, confirm that the new engine returns equivalent timing data. Generative does not return speech marks.
  • Lexicons: Polly's custom lexicons do not have a one-to-one equivalent in this comparison. Test pronunciations on each vendor.
  • Voices: voice names and styles differ between vendors. Re-audition the voices with your own scripts.
  • Limits: the synchronous and asynchronous character limits differ by vendor, so split long documents to fit each one.

Where OndaLabs fits

For a team that needs timed speech marks, AWS compliance terms or per-character API billing, Polly remains a reasonable choice, and this guide does not argue otherwise. OndaLabs is an AI voice studio for finished audio, not a replacement for Polly's API. Its published facts do not include API features, so use it for production work rather than as an API drop-in.

OndaLabs offers text to speech with 30 prebuilt voices usable in every supported language, plus dedicated pages for 37 languages. Its plans are credit-based: Onda Lite costs 1 credit per character and Onda Flash costs 1.5 credits per character on paid plans. The Starter plan costs $9 per month for 170,000 credits. At Onda Lite rates, that is 170,000 characters, which works out to about $52.94 per 1 million characters (derived). Plans also cover other tools, including image generation, so the comparison is not like-for-like. Review the pricing page before you decide.

Frequently asked questions

What is the best alternative to Amazon Polly?

There is no single best option. Google Standard and WaveNet match Polly Standard at $4 per 1 million characters, Azure Neural and OpenAI tts-1 sit near $15, and each vendor has different limits and voice sets. Choose by the job, your compliance needs and a listening test.

How much does Amazon Polly cost per million characters?

Standard costs $4, Neural $16, Long-Form $100 and Generative $30 per 1 million characters, based on AWS's pricing page checked on 2026-10-17. Free tiers differ by engine, and Neural, Long-Form and Generative free tiers last only the first 12 months.

Does Amazon Polly support voice cloning?

Polly's documentation describes Brand Voice, which AWS builds as a custom engagement for one organization. It does not describe self-serve voice cloning in the pages reviewed for this guide. Confirm with AWS if cloning is required.

Is Amazon Polly free?

Polly has a free tier. Standard includes 5 million characters per month, and Neural, Long-Form and Generative include smaller allowances for the first 12 months. New AWS customers from 2025-07-15 can also receive up to $200 in Free Tier credits.

Is Amazon Polly good for audiobooks or voice agents?

Polly Long-Form is designed for long-form narration, but it costs $100 per 1 million characters and runs only in US East. For voice agents, Polly Generative offers bidirectional streaming in eight regions, with an 8-transaction-per-second default. Test latency in your own region before you commit.

Which Polly alternative sounds the most natural?

No vendor's own documentation settles this, and third-party rankings are snapshots from one arena. Run a blind listening test with your own scripts, voices and listeners, and compare the results across the vendors on your shortlist.