What's New This Week
in AI

Mistral Large 4 is a frontier multimodal large language model with 1 trillion parameters and 49 billion active parameters. It handles text and image inputs, excels at coding, cybersecurity, and agentic workflows, and supports over 160 languages.

Embed 5 is Cohere's most powerful embedding family, delivering frontier retrieval quality across multimodal, multilingual, and financial documents. Available in Pro and Fast tiers that share one embedding space for flexible enterprise workflows.

OLMo-core 3 is an open-source training framework from the Allen Institute for AI that enables researchers to build and train large mixture-of-experts models at scale. It delivers 2.7x faster throughput and supports models beyond one trillion parameters.

SynthID Bio is a watermarking technology from Google DeepMind that embeds invisible, detectable signatures into AI-generated protein sequences and predicted 3D structures while preserving biological function for biosecurity and scientific integrity.

Praxis-1 is an open-weight world action model from Runway that turns large-scale video pretraining into real-world robot control. It works across different robot types and environments, learning from web video to reduce the need for costly robot demonstrations.

Dots by OpenAI are proactive AI agents powered by GPT-6 Astra that work continuously in the background on your goals. They connect to over 4,000 apps, conduct research, draft documents, write code, and learn your preferences over time while you stay in control.

GPT-6.1 Sol is OpenAI's mid-tier reasoning model that delivers near-Astra performance on coding, computer use, and professional tasks at one-fifth the cost. It features a 1.05M-token context window and improved factual accuracy.

Eleven v4 is ElevenLabs' most advanced text-to-speech model that interprets tone, pacing, emotion, and context to generate dramatic, tender, urgent, or conversational speech. It supports 90+ languages with voice cloning from just 10 seconds of audio.

Gemini 3.8 Live with Live Avatar pairs Google's speech-to-speech AI model with near real-time video generation to create dynamic, lip-synced avatars that can see, listen, and respond naturally across 97 languages for business applications.

GPT-6 Sol and Luna are large language models from OpenAI that bring GPT-6 Astra's intelligence to more affordable tiers. Sol handles complex coding and professional tasks, while Luna excels at high-volume clerical work like summarizing and data extraction.

Grok 4.7 is SpaceXAI's frontier large language model designed for software development, coding agents, and multi-hour professional tasks. It features a 500K token context window, improved self-verification, and strong performance on coding benchmarks.

Grok Voice Transcribe 2.0 is an AI-powered speech-to-text transcription model from xAI that converts audio to text with high accuracy. It handles batch files, URLs, and real-time streaming, with automatic language detection and speaker identification included.

SynthID Bio is a watermarking technology from Google DeepMind that embeds invisible, detectable signatures into AI-generated protein sequences and predicted 3D structures while preserving biological function for biosecurity and scientific integrity.

OLMo-core 3 is an open-source training framework from the Allen Institute for AI that enables researchers to build and train large mixture-of-experts models at scale. It delivers 2.7x faster throughput and supports models beyond one trillion parameters.

Dots by OpenAI are proactive AI agents powered by GPT-6 Astra that work continuously in the background on your goals. They connect to over 4,000 apps, conduct research, draft documents, write code, and learn your preferences over time while you stay in control.

Eleven v4 is ElevenLabs' most advanced text-to-speech model that interprets tone, pacing, emotion, and context to generate dramatic, tender, urgent, or conversational speech. It supports 90+ languages with voice cloning from just 10 seconds of audio.

Embed 5 is Cohere's most powerful embedding family, delivering frontier retrieval quality across multimodal, multilingual, and financial documents. Available in Pro and Fast tiers that share one embedding space for flexible enterprise workflows.

GPT-6.1 Sol is OpenAI's mid-tier reasoning model that delivers near-Astra performance on coding, computer use, and professional tasks at one-fifth the cost. It features a 1.05M-token context window and improved factual accuracy.

Praxis-1 is an open-weight world action model from Runway that turns large-scale video pretraining into real-world robot control. It works across different robot types and environments, learning from web video to reduce the need for costly robot demonstrations.

Generates AI-processed "undressed" images from photos. It removes clothes from pictures virtually.

Grok 4.7 is SpaceXAI's frontier large language model designed for software development, coding agents, and multi-hour professional tasks. It features a 500K token context window, improved self-verification, and strong performance on coding benchmarks.