How to Compare TTS Pricing Across Providers in 2026

By Gradium. Data as of September 2026.

Key takeaways

• This is the pricing cluster hub. Providers bill per character, per minute or in credits, so headline prices are not comparable until converted to one unit.
• Convert everything to cost per 1M characters, or per hour of finished audio. About 45,000 characters is one hour of speech at roughly 750 characters per minute.
• September 2026 rates per 1M characters: OpenAI tts-1 about $15, Deepgram Aura-2 $30, Gradium $35.90 to $57.80 by plan, ElevenLabs Flash v2.5 $50, ElevenLabs Eleven v3 $100.
• Speech-to-Text billing changes total cost per conversation more than the synthesis rate does. Gradium draws it from the same credit pool at 3 credits per second; ElevenLabs bills Scribe v2 separately at $0.22 per hour.
• Check three things no pricing page fronts: the concurrency ceiling on the plan you would actually buy, whether unused credits roll over, and whether a promotional rate has an end date.

TTS pricing is easy to misread. Providers use different billing units (characters, credits, minutes, tokens), bundle different features into the base price, and structure tiers in ways that make apples-to-apples comparison harder than it looks. A provider at $50/1M characters sounds cheaper than another at $340/month until you calculate what the monthly plan actually delivers.

This article covers how TTS pricing models work, what the headline price hides, and how to calculate the real cost of a TTS pipeline at your projected scale.

What pricing models do TTS APIs use?

Per-character pricing

Per-character pricing charges a fixed amount per million characters of text synthesized. It scales linearly with usage and is straightforward to project: estimate your monthly character volume, multiply by the rate, and you have your cost.

This model is used by OpenAI (tts-1 at about $15/1M, tts-1-hd at about $30/1M), Inworld (Realtime TTS-2 at $25/1M on demand down to $5/1M at enterprise volume, TTS-2 Flash at $15/1M down to $7/1M), Google Gemini 3.1 Flash TTS ($36.6/1M), Fish Audio S2 Pro ($15/1M), and Azure AI Speech HD 2.5 ($22/1M). ElevenLabs also publishes per-character equivalent rates for its models (Flash v2.5 and Eleven v3 Conversational at $50/1M, Eleven v3 and Multilingual v2 at $100/1M).

Deepgram Aura-2 uses per-character pricing at $0.030 per 1,000 characters ($30/1M), with a lower rate of $0.027/1,000 characters ($27/1M) at the Growth tier.

Credit-based subscriptions

Credit-based subscriptions give a monthly credit allocation that covers multiple services. Credits are consumed per request, with different services costing different amounts per credit.

Gradium uses this model. The conversion rate is straightforward: 1 character of TTS equals 1 credit, and 1 second of STT equals 3 credits. Both draw from the same monthly credit pool. A minute of TTS audio typically requires approximately 750 characters, which corresponds to 750 credits, though this varies by language and speaking pace.

Cartesia also uses a credit system: the Startup plan offers 1.25 million credits at $39/month (annual billing), and the Scale plan offers 8 million credits at $239/month (annual billing).

ElevenLabs uses a credit system internally but publishes per-character equivalent rates, with Flash v2.5 at 0.5 credits per character and Multilingual v2 at 1 credit per character on credit-based plans, though the rates vary by plan tier.

The key question with any credit-based system is what else those credits cover. A credit pool that includes both TTS and STT is economically different from one that covers TTS only.

Token-based pricing

OpenAI's gpt-4o-mini-tts is token-priced rather than per character, which OpenAI expresses as roughly $0.015 per minute of audio. That is a different structure from a per-character rate: it scales with output duration rather than with input length, so a slow, deliberate delivery of the same script costs more than a brisk one.

What does the headline price not tell you?

STT bundled vs separate

A voice agent pipeline requires both TTS (the agent speaking) and STT (transcribing what the user says). If your TTS provider does not include STT, you need a second vendor for transcription, which adds cost, integration work, and a second billing relationship.

Gradium includes streaming STT with semantic VAD in the same credit pool as TTS. 1 second of STT costs 3 credits, sharing the same monthly allocation as TTS. At the S plan ($43/month, 900k credits), this covers approximately 20 hours of TTS or 83 hours of STT, or a combination of the two.

ElevenLabs STT (Scribe v2) is a separate product with separate billing. Deepgram's STT and TTS are both available but billed separately. For voice agent pipelines, the difference between bundled and separate STT affects the total pipeline cost significantly, particularly at high STT volumes.

Voice cloning included vs gated

Voice cloning availability varies substantially between free and paid tiers across providers. On Gradium, Instant Voice Cloning (from 10 seconds of audio, up to 1,000 clones per month) is available from the free tier, with 5 clones included at $0/month. Professional Voice Cloning, a fine-tuned model for higher speaker fidelity, is available from the M plan (5 included) and the L plan (20 included).

On ElevenLabs, voice cloning requires a paid plan. On Cartesia, voice cloning is available on paid plans. Deepgram Aura-2 does not offer voice cloning. OpenAI's Voice Engine remains in limited preview as of 2026.

If voice cloning is part of your product requirements, factor in whether it is included at the entry tier or requires a plan upgrade.

Billing units: characters vs minutes vs tokens

The most common source of pricing confusion is that providers measure usage in different units. Characters and tokens are not the same. Audio minutes and character counts are not directly comparable without knowing your average words per minute and characters per word.

The conversion that matters for TTS: approximately 750 characters equals 1 minute of audio output (Gradium FAQ). Knowing this, you can convert per-character rates to per-minute equivalents for comparison. At $35.9/1M characters (Gradium L plan), 1 minute of audio costs approximately $0.027. At $50/1M (ElevenLabs Flash v2.5), 1 minute costs approximately $0.0375.

How do TTS prices compare across providers in 2026?

Sources: Official pricing pages (May to July 2026). Gradium pricing from gradium.ai/pricing. Competitor pricing verified at publication.

How do you calculate your real TTS cost at scale?

Step 1: estimate your monthly character volume

Start with your average response length in words and your expected monthly conversation volume.

A rough baseline from Gradium's documentation: one minute of TTS audio typically requires approximately 750 characters. Working backwards: if your average agent response is 50 words, that is approximately 300 characters per turn. At 100,000 turns per month, your TTS volume is approximately 30 million characters.

You can also estimate from audio minutes: multiply your expected monthly audio output hours by 60 (minutes) by 750 (characters per minute) to get a character count.

Step 2: account for what is included

Before comparing per-character rates, check what is included in the plan:

STT usage. If you need real-time transcription alongside TTS, calculate your expected STT volume separately. On Gradium, STT draws from the same credit pool at 3 credits per second of audio. On providers where STT is separate, add the cost of your STT provider to the TTS cost to get the real pipeline total.

Voice cloning. If your product uses a custom or cloned voice, confirm whether cloning is available at your target tier and whether there is a per-clone cost or a monthly limit. Gradium includes up to 1,000 Instant Voice Clones per month on paid plans.

Annual vs monthly billing. Gradium offers one month free on annual billing. Several other providers also discount annual plans. If you have committed budget, annual billing lowers the effective monthly rate.

Step 3: compare total pipeline cost

At 30 million TTS characters per month plus 300 hours of STT:

Gradium L plan ($1,615/month, 45M credits): 30M TTS characters consume 30M credits. 300 hours of STT is 300 × 3,600 seconds × 3 credits = 3.24M credits. Total: approximately 33.2M credits, within the 45M allocation of a single L plan. Full pipeline cost: $1,615/month, TTS and STT included, effective TTS rate $35.9/1M.

ElevenLabs Flash v2.5 ($50/1M): 30M chars = $1,500/month TTS only, plus Scribe v2 for 300 hours of STT at additional cost.

OpenAI TTS-1 ($15/1M): 30M chars = $450/month TTS only, no STT included, no voice cloning.

The lowest per-character rate is not always the lowest total pipeline cost. A provider with a higher per-character rate but bundled STT may cost less than a cheaper TTS provider paired with a separate STT service.

How does Gradium's pricing work?

Gradium's pricing page (gradium.ai/pricing) uses a single credit pool for all services. The current plans are:

Source: gradium.ai/pricing.

All paid plans include annual billing with one month free. The free plan requires no credit card and does not expire.

Per-character equivalents for TTS only (1 credit = 1 character):

• XS: $13 / 225,000 chars = ~$57.8/1M chars
• S: $43 / 900,000 chars = ~$47.8/1M chars
• M: $340 / 9,000,000 chars = ~$37.8/1M chars
• L: $1,615 / 45,000,000 chars = ~$35.9/1M chars

These rates apply to TTS only. When STT usage is included in the same credit pool, the effective TTS rate per dollar depends on how the credits are split between TTS and STT.

Startup Program. Seed-funded startups can apply for $2,000 or more in free credits, 6 months of full API access (M plan equivalent), direct engineering support, and early access to new models. Available at gradium.ai/blog/gradium-startup-program.

Glossary

Per-character pricing

A TTS billing model that charges a fixed amount per million characters of input text synthesized. Scales linearly with usage. Used by OpenAI, Inworld, Google, Fish Audio, Azure, and Deepgram. Comparable across providers after adjusting for what is included.

Credit-based subscription

A billing model providing a monthly credit allocation that covers multiple services. Gradium uses this model: 1 TTS character = 1 credit, 1 STT second = 3 credits. Both services draw from the same pool. Cartesia and ElevenLabs also use credit systems with different internal conversion rates.

Effective per-character rate

The per-character equivalent cost derived from a subscription plan. For Gradium, this ranges from $57.8/1M at the XS plan to $35.9/1M at the L plan, calculated by dividing the plan price by the character equivalent of the credit allocation.

Bundled STT

STT (Speech-to-Text) transcription included in the same plan and billing pool as TTS. Gradium includes streaming STT with semantic VAD at 3 credits per second in all plans. Providers where STT is a separate product (ElevenLabs Scribe v2, Deepgram billed separately) require additional budget for voice agent pipelines.

Instant Voice Clone

A voice cloning tier that creates a usable synthetic voice from a 10-second audio sample, immediately available for TTS generation. Gradium includes up to 1,000 Instant Voice Clones per month on paid plans and 5 on the free tier. Whether voice cloning is included in the base plan or gated behind higher tiers varies by provider.

Annual billing discount

A pricing option where committing to 12 months of a plan reduces the effective monthly rate. Gradium offers one month free on annual billing. Several other providers offer similar discounts.

Related guides

Part of the pricing and scale cluster. The pages it connects to:

• Best Text-to-Speech APIs 2026: the broad provider ranking, where these prices buy their quality.
• Best TTS API 2026: the short verdict page, one pick per use case.
• Best speech APIs 2026: Text-to-Speech and Speech-to-Text on one bill.
• ElevenLabs vs Cartesia vs OpenAI vs Gradium: the four-way head-to-head, with its own rate table.
• Best ElevenLabs alternatives 2026: voice quality against price across the field.
• How to choose a TTS API: the criteria framework, of which cost is the fifth.
• Best low-latency TTS APIs 2026: what concurrency limits cost in practice.
• How to estimate TTS costs at scale: a worked credits-and-concurrency example, with a provider comparison.

Beyond this topic

Price per character is only half the bill. Best speech APIs 2026 covers what Speech-to-Text adds, and WebSocket multiplexing covers the per-turn overhead that does not appear on any pricing page.

Gradium's free tier includes 45,000 credits per month with no credit card. API reference and quickstarts are at docs.gradium.ai. For enterprise evaluations, use our contact form.