# ElevenLabs introduced Flash with English and 32-language variants ElevenLabs introduced Flash on December 18, 2024, pairing low-latency text-to-speech with separate English-only v2 and 32-language v2.5 model variants. Canonical: https://rangoon.ai/insights/archive-elevenlabs-flash-december-2024/ Author: Rangoon Editorial (https://rangoon.ai/insights/editorial/) Published: 2026-10-24 Updated: 2026-10-24 Event date: 2024-12-18 Tags: [ElevenLabs](https://rangoon.ai/insights/tags/elevenlabs/), [speech synthesis](https://rangoon.ai/insights/tags/speech-synthesis/) Source note: Historical reporting: Source 0 is ElevenLabs’ dated Flash announcement. Source 1 is the currently published model reference, read for model identifiers, language coverage and configuration limits only; it does not establish December 2024 availability. Image: Flash variants from text to speech. Original Rangoon explanatory diagram: Source boundary: dated primary release reporting; original editorial analysis. No vendor interface, photograph, measurement result, or endorsement implied. ## Key takeaways - ElevenLabs dated Flash’s release to December 18, 2024 and named it for low-latency speech generation. - The release split Flash v2, which was English-only, from Flash v2.5, which ElevenLabs said supported 32 languages. - ElevenLabs attributed the 75 ms figure to generation and excluded application and network time; current documentation also warns that number normalization is disabled by default for v2.5. ## Flash arrived as a speed-oriented speech option [source 1](https://elevenlabs.io/blog/meet-flash) ElevenLabs published Meet Flash on December 18, 2024. The company described Flash as its newest text-to-speech model and said it generated speech in 75 milliseconds plus application and network latency. It recommended the model for low-latency conversational voice agents, said it could be used in the company’s Conversational AI platform, and named two API model identifiers: eleven_flash_v2 and eleven_flash_v2_5. This was a product-distribution event, not a general claim that a whole voice application would respond within 75 milliseconds. Speech generation is one stage in a system that may also include input capture, turn handling, model reasoning, network transit, playback and tool calls. The announcement supplies a dated vendor statement about the speech model’s intended role; a deployment needs its own end-to-end measurements. ## Language coverage was deliberately split by variant [source 1](https://elevenlabs.io/blog/meet-flash) [source 2](https://elevenlabs.io/docs/overview/models) The release made a concrete distinction that is easy to lose when Flash is discussed as one model family. Flash v2 was English-only, while Flash v2.5 supported 32 languages. ElevenLabs also said both variants cost one credit for every two characters at the time of the announcement. That structure matters for a product team choosing a speech path: a fast English model and a multilingual fast model have different language boundaries, even if they share the Flash name. The current model reference still lists eleven_flash_v2_5 separately and describes its language set as the Multilingual v2 languages plus Hungarian, Norwegian and Vietnamese. A later reference can clarify present identifiers and support, but it cannot prove every plan, price, API behavior or language quality available on the December release date. ## The launch framed speed as a tradeoff [source 1](https://elevenlabs.io/blog/meet-flash) [source 2](https://elevenlabs.io/docs/overview/models) ElevenLabs did not present Flash as a universally higher-fidelity replacement for its Turbo models. Its announcement said Flash had somewhat lower quality and emotional depth than Turbo while offering lower latency. It also reported blind tests by its human labelers against comparable ultra-low-latency models. Those are supplier-reported comparisons, not an independent benchmark and not a guarantee about a particular voice, language, text length or listening environment. The point of the release was therefore narrower and more useful: it exposed a model family for interactive speech workloads where response time has a high product cost. Selection still needs listening tests with the chosen voice, target languages, difficult names, punctuation, numbers and real response lengths. The provider’s category comparison does not settle those application-specific questions. ## Fast synthesis retained text-preparation limits [source 2](https://elevenlabs.io/docs/overview/models) The current ElevenLabs model page provides a technical caveat that helps explain the operational boundary. It says text normalization is disabled by default for Flash v2.5 to maintain its low-latency behavior, and calls out phone numbers, dates and currencies as cases that may be read in an unexpected way. It says Enterprise customers can set apply_text_normalization to on, while recommending that an upstream language model normalize text for low-latency or Agents Platform uses. This is context from a later, current document, not a launch feature claim. It shows why a historical statement about fast speech should not become a claim that all strings are spoken safely or clearly. The dated event establishes Flash’s release and vendor positioning; input transformation, plan entitlement, regional behavior and end-user comprehension remain separate work. ## Sources 1. [Official Flash release](https://elevenlabs.io/blog/meet-flash) 2. [ElevenLabs model reference](https://elevenlabs.io/docs/overview/models) ## Related reading - [ElevenLabs released Scribe as its first speech-to-text model](https://rangoon.ai/insights/archive-elevenlabs-scribe-february-2025/) - [ElevenLabs launched Eleven v3 as an alpha for expressive speech, with real-time use still out of scope](https://rangoon.ai/insights/archive-elevenlabs-eleven-v3-alpha-june-2025/) - [ElevenLabs launched Eleven Music with prompt-led song controls](https://rangoon.ai/insights/archive-elevenlabs-eleven-music-august-2025/) ## More information - [Documentation index](https://rangoon.ai/llms.txt) - [AI access and policies](https://rangoon.ai/ai/) - [Source repository](https://github.com/hypler-dev/rangoon)