AI Voice, Image & Video API Cost Calculator

Pick what you are generating and enter your monthly volume. The table then prices it on current models from ElevenLabs, OpenAI, Cartesia and Deepgram for speech, and from Black Forest Labs, Google and Runway for images and video. Each row shows the price as the provider lists it, the cost per 1,000 minutes of speech or per 1,000 generations, and the month's total. The cheapest row is highlighted unless two tie, which is why the speech table opens with no highlight: OpenAI's tts-1 and Deepgram's Aura-1 both charge $0.015 per 1,000 characters. Everything is calculated in your browser.

How each medium is priced

Every row reduces to the same arithmetic. Count the billable units in your month, multiply by the provider's rate, then scale to a thousand of your own units so that different meters can be compared:

monthly = units × rate, and per 1,000 = monthly ÷ volume × 1,000

For speech the unit is almost always the character, so the calculator first turns minutes into characters:

characters = minutes × characters per minute

Images are billed per image, or per megapixel for FLUX.2, at about one megapixel each. Video is billed per second, but only in the clip lengths a model offers, so each clip is rounded up to the next allowed length before it is multiplied by the per-second rate and the number of clips.

At the defaults, 10,000 minutes at 1,000 characters a minute come to 10,000,000 characters. Eleven v4 at $0.08 per 1,000 characters costs 10,000 × $0.08 = $800.00 a month, or $80.00 per 1,000 minutes. Cartesia sells Sonic-3.6 only through plans, so the calculator tries each one: Scale's $299 covers 8,000,000 characters and the other 2,000,000 cost $38 per million, another $76, for $375.00, against $442.75 on Startup and $648.50 on Pro. OpenAI's gpt-realtime-2.1-mini bills audio by the token, one per 50 milliseconds or 1,200 a minute. That is 12,000,000 audio tokens at $20 per million, $240.00, and the script and its transcript add 2,500,000 text tokens each way at four characters a token, $1.50 in and $6.00 out, for $247.50 in all.

Where per-unit media pricing came from

Speech was the first generative medium sold by the unit. When Amazon launched Polly on 30 November 2016, the announcement priced synthesis at $0.000004 a character, or about $0.004 a minute of generated audio, after five million free characters a month. That ratio, a thousand characters to the minute, is the default here. Two months earlier, on 8 September 2016, DeepMind had described WaveNet, a network that generates the raw waveform one sample at a time, at 16,000 or more samples a second. In blind listening tests it scored 4.21 out of 5 for US English, against 3.86 for Google's best concatenative system, 3.67 for its parametric one and 4.55 for recorded human speech. Neural voices have been sold by the character ever since.

Images followed with a different unit. OpenAI opened its DALL·E API as a public beta on 3 November 2022 at $0.02 per 1024 × 1024 image, $0.018 at 512 × 512 and $0.016 at 256 × 256, so a picture's price depended on its size rather than its prompt. Black Forest Labs, whose researchers had built VQGAN, Latent Diffusion and Stable Diffusion, released FLUX.1 on 1 August 2024 with its [pro] model offered only through an API. Its current models still bill by size, per tier or per megapixel.

Video arrived priced by the second. Runway described Gen-1, which edits existing footage to match a text or image prompt, in a paper submitted in February 2023, and opened an API for Gen-3 Alpha Turbo on 16 September 2024, first to select partners. OpenAI added Sora 2 and Sora 2 Pro to its API on 6 October 2025, billed per second of output.

The past year has been one of withdrawals. OpenAI removed the Sora models and its Videos API on 24 September 2026 and names no replacement. Google shut down every Imagen 4 model on the Gemini API on 17 August 2026, naming gemini-nano-banana-2.1 as the successor, and describes the dates it lists as the earliest possible ones. On 1 October 2026 OpenAI also announced that tts-1, tts-1-hd and gpt-4o-mini-tts leave the API on 6 January 2027, in favour of gpt-realtime-2.1-mini on the Realtime API. Neither shut-down service has a row here, and the two retiring tts models keep theirs until January.

Current prices

The tables below are generated from the same data file the calculator reads, so the two cannot drift apart. Prices are standard pay-as-you-go rates in US dollars from each provider's own pricing pages, verified in October 2026.

Speech

Images

Video

Cartesia is the one provider that sells credits only through monthly plans, which makes its price per minute depend on your volume. A plan's fee buys its allowance whether you use it or not, and its overage rate is higher than its allowance rate, so the effective price is lowest just as a plan fills: $37.38 per 1,000 minutes at exactly 8,000 minutes on Scale, but $63.07 at 776 minutes on Pro, just before Startup becomes the cheaper plan. Above Scale's allowance the price creeps back up towards $38.00 per 1,000 minutes.

Clip lengths matter as much as rates. Runway's Gen-4.5 and Gen-4 Turbo take any whole number of seconds from 2 to 10, Veo 3.1 only 4, 6 or 8, and FLUX 3 Video 5 to 20. Ask for 5 seconds and Veo bills 6, so 2,000 silent clips cost $2,400.00 a month on Veo 3.1 instead of the $2,000.00 its per-second rate suggests. Ask for 12 and only FLUX 3 Video can make the clip, at $4,080.00 a month.

Images and video at the defaults

Ten thousand images a month cost $90.00 on FLUX 3 Image Klein and $1,340.00 on Nano Banana Pro, a spread of almost fifteen times between models that all work at about one megapixel. Google's successor to Imagen, Nano Banana 2.1, sits near the cheap end at $336.00, and Runway's Gen-4 Image Turbo costs $200.00 but only works from one to three reference images.

For video, 2,000 eight-second clips without sound cost $800.00 on Gen-4 Turbo, which animates an image rather than starting from text. Veo 3.1 Fast costs $1,600.00, Gen-4.5 $1,920.00, FLUX 3 Video $2,720.00 and Veo 3.1 $3,200.00. Tick the sound box and the two Runway models drop out, because neither generates audio. Veo 3.1 Fast becomes the cheapest at $2,400.00, ahead of FLUX 3 Video, which lists one rate with sound included, and Veo 3.1 at $6,400.00.

What the estimate leaves out

This is a list-price estimate. Free allowances are left out: Deepgram's $200 starting credit, the characters included in ElevenLabs' free tier and Cartesia's free plan. So are committed and batch prices, such as Deepgram's Growth plan, about a tenth cheaper for text-to-speech with a year paid up front, and Google's Batch API at half price, and promotions such as Deepgram's offer to match spending on Flux TTS with credit up to $500 until 31 December 2026.

Generation also has costs a meter does not show. A discarded image or clip costs the same as a kept one, so if you keep one take in three, multiply your volume by three. Larger outputs cost more, as FLUX 3 Image does at $0.607 for 4k and Runway's Gen-4 Image at 8 credits for 1080p, and editing, upscaling and reference images bill at their own rates. A voice agent also needs speech recognition and a language model: the LLM API cost calculator prices the model, and the LLM token calculator explains the four-characters-a-token rule this page applies to the realtime row. Enterprise contracts, which most of these providers offer at high volumes, usually come in below list prices.

Prices change without notice and figures here are estimates, not quotes; this site is not affiliated with any provider named above. See the site disclaimer.

Frequently asked questions

How much does ElevenLabs cost per minute of speech?

ElevenLabs bills its API in US dollars per character, at the same rate on every plan: $0.08 per 1,000 characters for Eleven v4, v3 and Multilingual v2, and $0.04 for Eleven v4 Turbo, Flash v2.5 and v3 Conversational. ElevenLabs equates 1,000 characters with about a minute of audio, so 1,000 minutes cost about $80.00 on v4 and $40.00 on v4 Turbo.

What happened to the Sora API?

OpenAI removed the Videos API and the sora-2 and sora-2-pro models on 24 September 2026, six months after notifying developers on 24 March, and lists no replacement. Sora 2 had been billed per second of video, at $0.10 a second at 720p. The video table here prices Runway's Gen-4.5 and Gen-4 Turbo, Google's Veo 3.1 through Runway, and Black Forest Labs' FLUX 3 Video instead.

What replaced Google's Imagen API?

Google shut down Imagen 4, Imagen 4 Ultra and Imagen 4 Fast on the Gemini API on 17 August 2026 and names gemini-nano-banana-2.1 as the replacement. It costs $0.0336 per 1K image, or $33.60 per 1,000 images, and the calculator also lists Nano Banana 2 at $0.067 and Nano Banana Pro at $0.134 per image.

Is OpenAI retiring tts-1?

Yes. OpenAI announced on 1 October 2026 that tts-1, tts-1-hd and gpt-4o-mini-tts leave the API on 6 January 2027, and names gpt-realtime-2.1-mini on the Realtime API as the replacement. At 1,000 characters a minute that model costs $24.75 per 1,000 minutes, against $15.00 for tts-1 and $30.00 for tts-1-hd.

How many characters are in a minute of speech?

It depends on the voice's pace, so the calculator lets you set it. Amazon priced Polly in 2016 at $0.000004 a character, or about $0.004 a minute, which is 1,000 characters a minute, and ElevenLabs uses the same ratio. Cartesia says a minute takes 750 to 800 credits at one credit a character. At 750, Eleven v4 falls from $80.00 to $60.00 per 1,000 minutes.

Why does a 5-second Veo clip cost more than five seconds of video?

Veo 3.1 on Runway only makes clips of 4, 6 or 8 seconds, so a 5-second request is billed as 6 seconds: $1.20 instead of $1.00 at the silent rate of $0.20 a second. The calculator rounds every clip up to the next length each model allows, and marks a model as not offered when the clip is longer than it can make.

Sources

1. API pricing, ElevenLabs, 2026
2. Pricing, OpenAI API documentation, 2026
3. Deprecations, OpenAI API documentation, 2026
4. Realtime API costs, OpenAI API documentation, 2026
5. Pricing, Cartesia, 2026
6. Pricing, Deepgram, 2026
7. Release notes, Black Forest Labs, 2026
8. What are the costs associated with using your models?, Black Forest Labs help center, 2026
9. Gemini Developer API pricing, Google AI for Developers, 2026
10. Gemini deprecations, Google AI for Developers, 2026
11. API pricing, Runway API documentation, 2026
12. Amazon Polly – Text to Speech in 47 Voices and 24 Languages, AWS News Blog, 2016
13. WaveNet: A generative model for raw audio, Google DeepMind, 2016
14. Now anyone can build apps that use DALL-E 2 to generate images, TechCrunch, 2022
15. Announcing Black Forest Labs, Black Forest Labs, 2024
16. Structure and Content-Guided Video Synthesis with Diffusion Models, arXiv, 2023
17. Introducing the Runway API, Runway, 2024
18. Changelog, OpenAI API documentation, 2026