Docs / Billing & Credits

Complete Credit Rates: Exactly What Each Brand and Tier Charges

This page is the single authoritative price list for Soniva text-to-speech (character-based billing). Every figure comes directly from Soniva's backend pricing configuration — not rounded marketing numbers. Any deduction you see in the product can be reconciled against this table.

This page is the single authoritative price list for Soniva text-to-speech (character-based billing). Every figure comes directly from Soniva's backend pricing configuration — not rounded marketing numbers. Any deduction you see in the product can be reconciled against this table.

AI chat and translation are billed per token and have their own table: AI Model Credit Rates — the same model costs three different amounts on DeepSeek, Volcengine Ark and Alibaba Cloud Model Studio, so that page is grouped by platform.


Three Things to Understand First

1. What Credits Are

Soniva has exactly one billing unit: credits. No matter which brand's voice you use, every deduction comes from the same credit balance.

2. How Rates Are Set

They didn't come out of thin air — there's a single public formula:

Credits per 1,000 characters = that tier's official upstream unit price (USD per 1M characters) × 1.15

What the 1.15 is: Soniva adds 15% on top of official upstream prices to cover the cost of retrying failed upstream calls, USD/CNY exchange-rate fluctuation, and a small amount of service overhead.

The face value of a credit: 1,000 credits = the $1 that Soniva actually receives. Roughly 24% of what you pay when topping up is taken by payment-provider fees and local taxes, so $5 yields 3,800 credits (see Top-Up Rules). Stack the two together and the credits you buy cost roughly 1.5× the official upstream price — that is the entire premium for using Soniva instead of opening your own accounts across 8 platforms. There is no other hidden markup.

What this means: expensive brands drain your balance quickly, cheap ones slowly, and the difference comes entirely from the upstream brands' own pricing. Soniva never singles out a brand for an extra surcharge. The cheapest tier and the most expensive tier differ by roughly 42× — and that is precisely the point of a unified credit pool: with the same budget, you can either read a million characters with Tencent Cloud's premium voices, or read 24,000 characters with ElevenLabs' top-tier voices. The choice is yours.

3. Exchange-Rate Basis

  • CNY→USD is uniformly converted at 1 USD = 7.2 CNY (all domestic upstream vendors quote in CNY).
  • The "official upstream unit price" column is taken from each vendor's public price list, with a verification date and source link so you can cross-check it yourself.

Full Rate Table (Cheapest to Most Expensive)

BrandTier / ProductOfficial Upstream PriceApprox. USD per 1M charsCredits / 1,000 chars
Tencent CloudPremium Voice¥0.3 / 10,000 chars$4.175
Baidu CloudBasic Voice Library¥2.0 / 1,000 calls$4.635
Baidu CloudPremium Voice Library¥4.0 / 1,000 calls$9.2611
Baidu CloudUltimate Voice Library / LLM Voice Library¥4.5 / 1,000 calls$10.4212
Alibaba Cloud BailianCosyVoice v3.5-flash¥0.8 / 10,000 chars$11.1113
Alibaba Cloud BailianCosyVoice v3-flash / Sambert¥1 / 10,000 chars$13.8916
Microsoft AzureNeural TTS$15 / 1M chars$15.0017
Tencent CloudLLM Voice / Multi-Emotion LLM¥0.55–1.2 / 10,000 chars$16.6719
Alibaba Cloud BailianCosyVoice v3.5-plus¥1.5 / 10,000 chars$20.8324
Alibaba Cloud BailianCosyVoice v1 / v2 / v3-plus¥2 / 10,000 chars$27.7832
Doubao SpeechSpeech Synthesis Model 2.0¥3 / 10,000 chars$41.6748
CartesiaSonic~1 credit / char$50.0057
MiniMaxTurbo (speech-2.8 / 2.6-turbo)¥2 / 10,000 chars (CJK chars count ×2)$55.5664
Doubao SpeechLLM Speech Synthesis 1.0¥5 / 10,000 chars$69.4480
Tencent CloudSuper-Natural LLM Voice¥4.9–6.5 / 10,000 chars$90.28104
ElevenLabsFlash / Turbo v2.5≈ $91 / 1M chars via Creator plan$91.00105
MiniMaxHD (speech-2.8 / 2.6-hd)¥3.5 / 10,000 chars (CJK chars count ×2)$97.22112
ElevenLabsMultilingual v2 / v3≈ $182 / 1M chars via Creator plan$182.00209
Doubao SpeechSpeech Synthesis Small Model (BV voices)¥5.5 / 1,000 callsPay-per-call, see belowPer call

Upstream prices take the conservative end (the more expensive one). For example, Tencent Cloud's LLM Voice ranges ¥0.55–1.2 / 10,000 chars officially, and this table uses ¥1.2; Doubao and Baidu also have cheaper prepaid bundles, and this table uses the more expensive postpaid price. The reasoning: it's better to start higher and adjust down once real bills arrive than to underprice first and raise prices later.


Three Billing Details You Must Know

Detail 1: MiniMax counts a CJK character as 2 characters

MiniMax's official billing rule: "1 CJK character counts as 2 billable characters; letters, punctuation, and spaces each count as 1." Every other brand counts 1 CJK character as 1 character.

The MiniMax rows above already have the ×2 factored in. Estimated for pure-Chinese content, the consumption rate shown in the product also uses this Chinese base case (worst-case display). So if you read pure English with MiniMax, your actual deduction is roughly half the figure in the table (Turbo ≈ 32, HD ≈ 56). It is the only per-character brand where "English is half the price of Chinese."

See the MiniMax guide.

Detail 2: Baidu bills per call, not per character

Baidu's official rule: every 120 GBK bytes counts as 1 call; anything under 120 bytes also counts as 1 (rounded up). Under GBK encoding, 1 CJK character/punctuation mark takes 2 bytes, and 1 letter/digit takes 1 byte.

In practice:

  • Chinese: 120 bytes ≈ 60 CJK characters = 1 call
  • English: 120 bytes ≈ 120 letters = 1 call

So the "credits / 1,000 chars" for Baidu's three rows above is an equivalence estimated for pure Chinese. When reading English, the same word count takes fewer bytes, so the actual deduction is roughly half of the Chinese figure.

Also, because billing rounds up, very short text (e.g., a single sentence) is still charged as at least 1 call — the implied unit price can look higher than the table. That's an inherent property of per-call billing, not an overcharge.

Detail 3: Doubao's small model (BV voices) is also per-call

Doubao Speech's "small model" product line (voice IDs beginning with BV, e.g. BV700_streaming) is billed by the upstream vendor per call: ¥5.5 / 1,000 calls, with a per-request text cap of about 300 CJK characters.

Soniva automatically splits long text into paragraphs of up to ~280 characters before sending, counts each paragraph as one call, and deducts a flat 1 credit per call. So:

  • Reading 280 chars or fewer → 1 credit
  • Reading 1,000 chars → ~4 credits
  • Reading 10,000 chars → ~36 credits

For short-text, high-frequency reading scenarios, this tier is usually the cheapest of all cloud voices — roughly 4 credits / 1,000 chars, even cheaper than Tencent's Premium tier (5).

See the Doubao Speech guide.


What Doesn't Consume Credits

These use cases deduct nothing and don't require a subscription:

Use caseDescription
System built-in voice readingUses your device/browser's own reading engine (Apple voices on Mac/iPhone, Microsoft voices on Windows, Google voices on Android/Chrome). Completely free and unlimited, but the voice is decided by your system — you can't choose a cloud AI voice.
Bring Your Own Key (BYOK)Enter an API key you applied for on platforms like Alibaba Cloud, Volcano Engine, or ElevenLabs in Settings, and requests go directly through your own account. The upstream platform bills you directly; Soniva deducts no credits.
Free previewAlibaba Cloud / Doubao / MiniMax / Tencent Cloud / Baidu / Microsoft each provide 1 representative voice + 1 fixed script. You can play it without logging in. Limit: 5 times per brand per day (counted by IP), and the script can't be changed.
ElevenLabs / Cartesia official previewsThese two brands' voice lists come with official pre-made preview audio that plays directly without going through Soniva's backend, so there's no separate whitelist preview.
Translation and AI chat inside the free allowance1,000 chars/day logged out, 5,000 chars/day logged in without a subscription — that part runs on a character quota and costs no credits. Beyond the free tier, subscribers are charged credits at the model's token price — see the AI Model Credit Rates.

Will This Table Change

Yes, but changes are traceable:

  • When an upstream brand adjusts prices, Soniva updates this table and refreshes the "verification date" on this page.
  • The 1.15 coefficient is Soniva's only pricing lever. If payment-provider fees or exchange rates shift significantly in the future, this number may be adjusted — with prior notice. Credits you've already purchased are never changed; what changes is how many credits a thousand characters cost.
  • What won't happen: a brand being quietly surcharged on its own. Rates are always "official upstream price × the same coefficient," applied equally to every brand.

Data Sources

BrandOfficial Price PageDate Verified
Baidu Cloudhttps://cloud.baidu.com/doc/SPEECH/s/Ql9misjot2026-08-18
Tencent Cloudhttps://cloud.tencent.com/document/product/1073/341122026-08-05
Microsoft Azurehttps://azure.microsoft.com/en-us/pricing/details/speech/2026-08-17
Alibaba Cloud Bailianhttps://help.aliyun.com/zh/model-studio/model-pricing2026-08-17
Doubao Speech (Volcano Engine)https://docs.volcengine.com/docs/6561/13593702026-08-19
MiniMaxhttps://platform.minimaxi.com2026-08-17
ElevenLabshttps://elevenlabs.io/zh/pricing2026-08-17
Cartesiahttps://www.cartesia.ai/pricing2026-08-06

Was this article helpful?

Anonymous feedback is enabled when you allow analytics and the service is available. You can also report an issue by email.

Report an issue by email

The email includes this article. You can also copy the details and send them to soniva.cde@gmail.com