Ad
Short

A new open source voice model called Kokoro just landed on HuggingFace, and early tests show it can generate voices that rival commercial services like Eleven Labs. The model packs 82 million parameters under the hood, and is on the first place in the TTS Spaces Arena. The model is trained on less than 100 hours of audio data, supporting just American and British English for now. Users can currently choose from 10 different voices. While the model shows promise, it does have its limitations. Unlike some commercial alternatives, it can't clone voices, and there aren't any plans to add support for other languages yet. For developers interested in using Kokoro, the inference code is available under an MIT license, while the model itself uses an Apache 2.0 license.

Ad
Ad
Short

French AI startup Mistral has released Codestral 25.01, an updated version of its code generation model. According to Mistral, the new model generates and completes code twice as fast as its predecessor while supporting over 80 programming languages. Benchmarks indicate that Codestral 25.01 outperforms other code models with fewer than 100 billion parameters on fill-in-the-middle tasks. It currently ranks second in LMsys' Copilot Arena, behind Sonnet 3.5 and Deepseek V2.5. Developers can access the model through Mistral AI's IDE partners and the APIs of Mistral, Google, Microsoft, and Amazon. For enterprise customers, the startup Continue offers local deployment options. A free trial of Codestral is available in VS Code or JetBrains.

Ad
Ad
Ad
Ad
Google News