Mistral releases Voxtral, open-weight TTS model that clones voices from 3 seconds of audio
Mistral has released Voxtral TTS, a 4-billion-parameter text-to-speech model that can clone voices from just three seconds of reference audio across nine languages. The model delivers 70ms latency for typical 10-second samples and outperformed ElevenLabs Flash v2.5 in naturalness tests. Voxtral is available via API at $0.016 per 1,000 characters and as open-weights on Hugging Face.
Mistral Releases Voxtral: Open-Weight TTS Model with Voice Cloning from 3-Second Samples
Mistral has released Voxtral TTS, its first text-to-speech model, positioning it as a compact alternative to closed proprietary systems. The model contains 4 billion parameters and supports nine languages: German, English, French, Spanish, and five others.
Key Technical Specifications
Voxtral's standout capability is voice cloning from minimal audio. The model requires just three seconds of reference audio to adapt to and replicate new voices, with support for emotionally expressive speech synthesis. Latency benchmarks show 70 milliseconds for a typical configuration processing 10-second speech samples with 500 characters of input text.
The model operates across a broader linguistic range than many competing TTS systems, though Mistral has not specified the complete language list beyond the four named examples.
Performance vs. Competitors
In human evaluation tests, Voxtral TTS scored higher on naturalness compared to ElevenLabs Flash v2.5 at comparable response times. However, this comparison has a timing caveat: ElevenLabs subsequently released version 3, which was not included in Mistral's evaluation. This means the benchmark reflects performance against a prior-generation ElevenLabs model rather than current-generation alternatives.
Availability and Pricing
Mistral offers three access paths for Voxtral TTS:
- API access: $0.016 per 1,000 characters
- Mistral Studio: Web-based testing interface
- Open-weights version: Available on Hugging Face for local deployment and fine-tuning
The open-weights release represents a departure from Mistral's approach with some of its larger language models, giving developers the ability to run Voxtral locally without relying on the company's infrastructure.
What This Means
Voxtral establishes Mistral as a competitor in the TTS market beyond its core language modeling business. The 4-billion-parameter size makes it accessible for resource-constrained deployments—substantially smaller than many alternatives—while the open-weights availability appeals to enterprises avoiding vendor lock-in. The three-second voice cloning threshold is practically significant, reducing friction for users who need quick voice adaptation. The API pricing at $0.016 per 1,000 characters is competitive but not a market undercut; comparison requires converting to per-token equivalents based on language-specific tokenization rates. The main strategic value lies in the open-source option, which appeals to builders wanting fine-tuning and deployment flexibility that proprietary APIs don't provide.
Related Articles
Mistral Releases Shieldstral, a 3B Open-Weights Safety Classifier That Matches Models 7x Its Size
Mistral has released Shieldstral, a 3B open-weights safety classifier that reframes content moderation as a policy-adaptive question-answering task. The model claims to match or outperform guard models up to 7x its size on text safety and multimodal benchmarks, and runs on a single 16GB GPU.
Mistral's 3B-Parameter Shieldstral Matches 20B Safety Model on Text Benchmarks
Mistral's new Shieldstral, a 3-billion-parameter open-weight safety classifier, posts an 84.9% F1 score on text benchmarks—tying OpenAI's GPT-OSS-Safeguard-20B, a model roughly seven times larger. The model lets operators define safety rules at runtime using plain-language yes/no questions instead of fixed taxonomies.
Mistral AI Releases Shieldstral-1.0-3B, a 3B-Parameter Policy-Adaptive Safety Classifier
Mistral AI has released Shieldstral-1.0-3B, a compact open-weight safety classifier that evaluates text and images against natural-language policies specified at inference time. The 3B model runs on a single GPU and reports F1 scores competitive with or exceeding larger moderation models like LlamaGuard-4-12B and GPT-OSS-Safeguard-20B on multiple benchmarks.
Alibaba Unveils Qwen3.8-Max, a 2.4T-Parameter Open-Weight Model for Coding and Agentic Work
Alibaba announced Qwen3.8-Max, a 2.4T-parameter flagship model targeting coding and long-horizon agentic work, with open weights promised for next week alongside Qwen3.8-27B. The model posted strong third-party benchmark results, ranking #4 in Frontend Code Arena and matching Claude Opus 4.7 on the Vals Index at roughly 2.3x lower cost.
Comments
Loading...