There are so many origin stories of punk that no single history can count as definitive. But there’s also no disputing its roots in the New York poetry scene from which Patti Smith emerged in the 1960...

A 27M-parameter model just outperformed giants like DeepSeek R1, o3-mini, and Claude 3.7 on reasoning tasks

Mistral, a French AI startup, launched Devstral 2, a new suite of models optimized for software engineering, alongside the Vibe CLI. The release includes two models: a 123-billion parameter model (Devstral 2) and a smaller 24B model (Devstral Small 2). Devstral 2, designed for agentic software development, boasts a 256K-token context window and impressive SWE-bench performance. Devstral Small 2, Apache 2.0 licensed, is ideal for local, private use and runs on a single laptop. Vibe CLI, a terminal-native agent, allows direct interaction with the models, enhancing developer workflow. Licensing is innovative: Devstral Small 2 uses Apache 2.0, while Devstral 2 has a modified MIT license limiting use for companies exceeding $20 million in monthly revenue. This licensing approach offers flexibility but also raises questions for enterprise adoption. Devstral Small 2 provides a strong, private alternative for enterprises. Mistral's strategy builds upon its Codestral and Mistral 3 releases, targeting a developer-focused, localized AI approach. The company aims for "distributed intelligence", offering various paths for adoption depending on user needs.

Handy is a cross platform, open-source, speech-to-text application for your computer