Voxtral | ai

Speech models of public speaking open speaking.

The expression of the installation of the limit with the opening code.

Voice: Original user interface.

The sound was the first interface of humanity - before long desk or writes, allowed to share ideas, coordinate and build relationships.While digital systems are more able, the sound returns than the nicest form of humanity interaction

Nevertheless, today's systems remain irrelevant, property and very fragile for use in the real world.It requires tools with exceptional transcription, deep concept, multilingual liquid and light, flexible placement to cancel this gap.

We posted the Voxral models to accelerate this future.The burning papers of the situation are available in two sizes - APLAINE to complication of apps.We have turned both models available on our API, and we also provided an end point for transcription only that provides cost efficiency without the route.

Open, affordable and production ready for everyone.

Until recently, gaining a true speech intelligence in production means choosing between two compromises:

Open source-source-source-resources with a useful power resource and useful limit.

The powerful writing spaces in a comprehensive form of understanding;

Vottrat BRID BRIG IZI.

The two voxtrales models go beyond transcription with options that include:

Long-term connection in the format: In order to copy 1 to 30 kilometers between 32 km, or 40 minutes

Kiguk-a and A.

Natural nature: Direct Language Discovery and High Performance in the most -used languages in the world (English, Spanish, French, Portuguese, Hindi, German, the Netherlands, Italy, calling several), helping teams to service the audience in one system with one system

Calling the action directly from the voice: It authorizes the activity of a user, translating the systems of speech, the System Refuse Orders Refund Order Badgerokeless.

Very capable in the text: Keep the functions of the understanding of text of your language model background model, small 3.1

These possibilities are best for contact in the real world and create low actions such as wrinkles, answers, analysis and observations.In the case of sensitive events that are sensitive to the cost of using the Vocstinal mini, the transcripted transcript exceeds the price of the whisper open.In the case of premium vocer -events, the writer's performance also applies to levanlops, which is less than half the price.

We value it in different English and multilingual criteria to assess the vocal transcription.For each project, we report the average error rate (lower than better) languages.

Vincal overforms ovegramonly, current major open string talk-crykver model.This GPT -4 is flashes on all deeds and both applications and receives the highest multilingual abilities.

When evaluated in Fleurs, smaller VOXTRALS are whispered on each task, acquiring modern performances in a number of European languages.

A brief form, livcrisclock, livcriscok, livcrisch, GGPIE, VicPPPEE - 4, SyntitkPPE

EN long -term: income -21 10 m, income -22 10 -m

Mozilla olu dara ụda 15.1: English, French, French, German, Spanish, Italian, Portuguese, Dutch, Dutch, Dutch, Dindi

Bonarams: English, French, German, Spanish, Portuguese, Duti, Arabue

Voxtral Small and Mini have the ability to answer questions directly from words or by providing a text -based and text reminder.To evaluate the ability to understand the sound, we create the general versions of the three tasks of understanding the common text.Translation of words on the benchmarks Fleurs Infusion.

VOXTRAL SMALL is competitive in GPT-4o-Mini and Gemini 2.5 blinking in all tasks, achieving sophisticated performance in speech translation.

VaxT retains the authority of their tongue, helping to use it as filling for the Ministry of Ministry 3.1 individuals individually.

Try it for free

Whether you create a prototype at Laptop, use personal loads in place or adjust production in the cloud.

Download and in local: all Walksters (24b) and Mini Walkster (3B) is available to download on the belt

Try API. Combine the border speech intelligence with a single API call.The price starts with $ 0.001 at $ 0.001 per minute.Take a look at our documents.

Try this at Le Chat: Try VOXTRAL in the voice mode of Le Chat (go out for all users in the coming weeks) - on the site or cellphone.Record or raise sound, get transcripts, ask questions or create summaries.

Huihui Huihui.

We also function functionality for VoxTal dinioted for the undertaken with the highest, upper-scale security scaling, or sample.s have considered it, please contact us:

Private layout on the production scale: Our solutions team can help you set a VXTRAL production scale for conclusions within your infrastructure.This is ideal for used cases in industrial areas with rigorous data protection requirements. Quantitative structures are tailored to production infiltration and cost effectiveness.

Specific Development Functions in the field: Collaborate with our AI team applies to adapting the Vaxtral in specialized context, as legal, medical, customer support or internal knowledge basis, improving the accuracy of the case of use.

Advanced context: We invite design partners to identify speakers to meet a wide range of needs outside the box, detect emotions, develop support for advanced steps and more context windows.

Support for Supreme Code: Access to technical resources and transactions to include pure voltran and employment activities or products.

We are registered with your friends with your friends VoxTral and Invorl Ttse Contal on Wednesday, sign up here.

We will be more fulfilling in the next month.In addition to our speech understanding, we will also support it quickly.

Labeling audio as age and emotion

Out -of -speech

We are happy to see what you are building with a weak.

VAT, BérelüNK!

The release of our volunteers of our watchpostal models is a significant step, but our journey is a long way from the most natural, attractive research and engineers to share our mall.

If you are interested in our mission in our mission to demand artificial intelligence, we welcomes your requests to join your team!

Put in contact.

To decrease the reafeation of the right of law's vocational models.