Eleven v4 tops the voice rankings at 5x Gemini's price

On Monday morning @ElevenLabs announced Eleven v4 and Eleven v4 Turbo, calling them "our fastest and most emotive voice models yet." Twenty-six minutes later, @ArtificialAnlys posted the independent numbers. Eleven v4 took first place on its Provider Voice TTS Arena, where listeners vote blind between voices.

That thread also listed prices, which ElevenLabs' own launch posts left out. If you're thinking about putting an AI voice on your phone line, the price line matters more than the Elo score.

What ElevenLabs shipped

Two models. Eleven v4 is the full version for narration, video and anything where delivery matters. Eleven v4 Turbo is the fast one built for live conversation, and on its voice agents page ElevenLabs puts Turbo's median response time at about 100 milliseconds. Both handle more than 90 languages, up from 70 in the last version, and both are available now in the company's agent builder and API.

The headline feature is control. You can write direction straight into the script with tags like [laughs] or [light rain], or describe how a line should sound, and the model acts it out. According to ElevenLabs' launch post, a voice recorded in one language can now speak another and still sound like the same person. Instant voice clones need only 10 seconds of audio.

What the benchmark says

@ArtificialAnlys had Eleven v4 at an Elo of 1,319, ahead of Cartesia's Sonic 3.6 at 1,276 and Google's Gemini 3.8 Flash TTS at 1,267. It ranked first in all four listener categories, customer service among them.

Two more results from that thread are worth knowing:

• On pronunciation robustness, the test of whether a model reads tricky words, numbers and names correctly, Eleven v4 scored 91.7 percent, the highest the group says it has measured.
• When every model used the same custom voice, Eleven v4 came second, behind Alibaba's Qwen-Audio-3.1-TTS-Plus.

That second result is easy to miss. My read is that some of ElevenLabs' lead comes from its stock voices, not only the engine underneath. Once every model reads in the same custom voice, it slips to second.

What it costs

Artificial Analysis lists Eleven v4 at $80 per million characters. Sonic 3.6 is $49 and Gemini 3.8 Flash TTS is $16.49. So the top-ranked voice costs close to five times what the third-place voice does, and the Elo gap between them is 52 points.

Here's my own rough math, not a published figure. A three-minute call where the agent does half the talking works out to something like 1,300 characters of speech. At API rates that's about ten cents per call on Eleven v4 and about two cents on Gemini. Over 1,000 calls a month, that's roughly $105 against $22 for the voice alone, before the language model, the phone line and the platform fee. Packaged agent products bill differently from raw API rates, so treat this as a sense of scale.

My read

For most small businesses, the voice isn't the part that fails. Callers hang up when the agent can't book the slot, doesn't know the service area, or misreads the address. The 91.7 percent pronunciation score speaks to the last one, and it's a real reason to pay more if your callers say street names, drug names or part numbers all day.

For a salon confirming Thursday appointments, the cheaper voices are now good enough that the premium is hard to justify. We wrote last week about Gemini Live Avatar and Qwen's 95 percent price cut. The cheap end keeps getting cheaper and the expensive end keeps getting better. You get to pick where you sit, and the best-known name isn't automatically the right pick.

One caution on the 10-second clone. It makes cloning a real staff member's voice almost effortless. Get written consent first, and read our post on Gemini's voice cloning and Illinois law before you do it here.

What this means for your business

Choose the voice last. First decide what the calls are for, what the agent can look up and book, and when it hands off to a person. Then test two or three voices on your real scripts, with your real street names, and pick the cheapest one your customers don't mind.

If you want help working out which calls are worth automating in the first place, New Face Design's free process audit is a good place to start. We'll map where your phone time goes and tell you honestly whether a voice agent pays for itself.