🎙️ Mistral launches Voxtral Open Source Speech Models 🎙️
High level speech understanding models available in 2 variants

Mistral AI has released Voxtral, a family of open-source speech understanding models that go beyond simple transcription to offer deep semantic understanding and multilingual capabilities. Available in two sizes—24B for production and 3B for edge deployment—both models are released under Apache 2.0 license and outperform OpenAI Whisper while costing less than half of comparable APIs. Voxtral handles 30-40 minute audio files with 32k token context, supports built-in Q&A and summarization, includes automatic language detection across major languages, and enables direct function calling from voice commands. The models retain text understanding capabilities from their Mistral Small 3.1 backbone, making them suitable for real-world applications requiring both transcription accuracy and semantic comprehension.

💻 Amazon releases Kiro, Claude-Powered IDE Rival 💻
New agentic development environment challenges Windsurf and Codex

Amazon has unveiled Kiro, a new agentic integrated development environment powered by Claude Sonnet 3.7 and 4.0 to compete with Windsurf and OpenAI Codex. The spec-driven development tool transforms high-level prompts into structured software requirements, including user stories, technical design documents, and task lists. Kiro features agent hooks for automation triggers like regenerating tests, updating documentation, and running security scans, while supporting Model Context Protocol (MCP) for external system integration. Built on Code OSS (Visual Studio Code's open-source foundation), Kiro runs on macOS, Windows, and Linux with free preview access (50 interactions per month) and paid tiers starting at $19.

🎬 Runway launches Act-Two Character Animation Tool 🎬
Enables realistic character animation using driving performance videos

Runway ML has introduced Act-Two, a groundbreaking character animation tool that allows users to animate characters using driving performance videos. The feature enables realistic motion transfer by providing a driving performance video and a character reference (image or video), automatically adding environmental motion to input images while maintaining gesture control capabilities. Act-Two works with Gen-4 Video model and supports up to 30-second duration with 24fps frame rate, offering multiple output resolutions from 720p to 1584x672 pixels. The tool features two input modes: character images that automatically add environmental motion and allow gesture control, while character videos retain subject environment and camera motion but disable gesture control.

Currently available for Enterprise and CPP accounts with phased rollout, Act-Two costs 5 credits per second with a 3-second minimum.

INTERESTING TO KNOW

🥽 ByteDance develops XR Glasses to Challenge Meta 🥽

ByteDance, TikTok's parent company, is reportedly developing mixed reality glasses codenamed "Swan" through its Pico VR division, directly challenging Meta's Project Orion. The lightweight goggles will weigh approximately 0.28 pounds and feature specialized chips to process sensor data and minimize latency between AR visuals and physical movements. Similar to Meta's approach, the device will use a connected puck (either wireless or wired) to offload processing power and reduce headset weight. This follows the canceled Pico 5 launch in 2023, marking ByteDance's pivot toward ultralight XR devices. The development comes as the industry shifts away from bulky VR headsets toward glasses-form factors, with competitors like Meta, Snap, Apple, and Xreal all racing to launch similar products. However, potential US market access remains uncertain due to ongoing TikTok ban concerns.

🚀 Mira Murati's Thinking Machines raises $2B at $12B Valuation 🚀

Mira Murati's AI startup Thinking Machines Lab has officially closed a $2 billion seed round led by Andreessen Horowitz, valuing the company at $12 billion. The deal includes participation from Nvidia, Accel, ServiceNow, CISCO, AMD, and Jane Street, marking one of the largest seed rounds in Silicon Valley history. The less than year-old startup has yet to reveal its products but promises to unveil its first product with significant open source components in the "next couple months." Murati has attracted former OpenAI colleagues including John Schulman, Barret Zoph, and Luke Metz to the venture. The massive funding represents investor appetite for promising AI labs and gives Murati enough resources to train frontier AI models and compete with established players like OpenAI, Anthropic, and Google DeepMind.