Enterprise

ElevenLabs ships Eleven v4 and a sub-100ms Turbo variant, with a two-week API discount aimed at small teams

A 10-second instant voice clone, 90-language coverage, and a promotional API price through Oct. 12 push branded voice agents within reach of founder-led sales and support teams.

Photo: Unsplash / Petr Macháček — Headset resting on a desk beside a laptop, evoking a customer-support or sales workstation

ElevenLabs launched Eleven v4 and a lower-latency Eleven v4 Turbo variant on September 28, 2026, and paired the release with a two-week API discount clearly aimed at pulling small teams onto the platform before pricing normalizes on Oct. 12. Through that date, v4 runs at $22 per million characters and Turbo at $11.

Turbo’s headline number is a median inference latency of roughly 100ms, with median time-to-first-speech around 150ms over WebSocket streaming. ElevenLabs’ own benchmarks put Cartesia Sonic 3.6 at 262ms on the same measure and OpenAI’s GPT-4o mini TTS at 814ms. Below about 200ms is the threshold where a synthetic voice stops sounding like it’s buffering and starts sounding like it’s talking back.

Language coverage moves from 70 in v3 to more than 90, with the company flagging the biggest quality jumps in Japanese, Brazilian Portuguese, Mandarin, and Cantonese. Per TechCrunch, v4 also handles confrontations, escalations, and holds differently, and can start generating audio as soon as the LLM behind it begins producing tokens.

The 10-second instant voice clone claim circulating from the launch thread deserves a caveat: RuntimeWire notes that ElevenLabs’ own v4 documentation says an Instant Voice Clone generally uses one to two minutes of source audio.

Deployment evidence is thin but pointed. Mikael Myhrberg, automation lead at PostNord, reports a 47% drop in speech time-to-first-byte across nine markets and 3,100 live calls after rolling Turbo into production, with no regression in resolution rate, quality, reliability, or cost. Oscar Daniels, head of credit building products at Spring Financial, framed it more bluntly: “this is the point where building stops feeling like an experiment.”

The competitive read-out matches. Artificial Analysis’ Provider Voice Arena ranked v4 #1 for September 2026 at Elo 1,319 across 1,674 samples, and ElevenLabs cites blind tests where roughly 75% of listeners preferred v4 over Cartesia, Inworld TTS-2, and two Gemini TTS variants.

Unite.AI notes a free tier of 10,000 credits per month, roughly ten minutes of audio, with paid plans starting at $6. TechCrunch pegs the run rate at over $600 million annualized, up from roughly $330 million at the start of 2026, following a $500 million Sequoia round earlier this year at an $11 billion valuation. The pricing window isn’t charity; it’s customer acquisition timed to a leaderboard moment.

Sources