AI News Hub

Open Source AI

Mistral Large 3 Released as Open Weight with Commercial License

Mistral AI releases Mistral Large 3, a 123B parameter dense model with open weights and a permissive commercial license. The model excels at European language tasks and features built-in function calling and JSON mode.

mistral large-3 open-weight +2

Feb 10, 2026 · Mistral

Read

Video Models

Pika 2.0 Launches Director Mode for Scene-by-Scene Control

Pika releases version 2.0 with Director Mode, giving creators scene-by-scene control over video generation including camera movements, character blocking, and timing. The feature makes professional video direction accessible to everyone.

pika video-generation director-mode +2

Feb 08, 2026 · Pika

Read

Text Models

Cohere Command A Optimized for Enterprise RAG Workflows

Cohere releases Command A, a language model specifically optimized for enterprise retrieval-augmented generation workflows. The model excels at grounded responses with built-in citation generation and hallucination reduction.

cohere rag enterprise +2

Feb 08, 2026 · Cohere

Read

Image Models

Adobe Firefly 3.0 Integrates Directly Into Creative Cloud Suite

Adobe launches Firefly 3.0 with deep integration across Photoshop, Illustrator, and Premiere Pro. The update brings structure-aware generation, brand kit consistency, and commercially safe AI image creation to professional workflows.

adobe firefly creative-cloud +2

Feb 07, 2026 · Adobe

Read

AI Agents & Tools

LangChain 1.0 Reaches Stable Release After Two Years

LangChain officially releases version 1.0 after two years of development, establishing a stable API for building LLM-powered applications. The release includes LangGraph for agent orchestration and LangSmith for production monitoring.

langchain langgraph langsmith +2

Feb 07, 2026 · LangChain

Read

AI Coding Agents

Amazon Q Developer Reaches General Availability with Enterprise Features

Amazon Web Services announces general availability of Amazon Q Developer with enterprise-grade features including private code customization, security scanning, and seamless integration with AWS infrastructure services.

amazon aws q-developer +2

Feb 06, 2026 · Amazon

Read

Open Source AI

Hugging Face Launches Open LLM Leaderboard v3 with New Metrics

Hugging Face unveils version 3 of its Open LLM Leaderboard with new evaluation metrics including real-world task completion, instruction following accuracy, and safety benchmarks. The update addresses criticism of benchmark gaming.

hugging-face leaderboard benchmarks +2

Feb 06, 2026 · Hugging Face

Read

Text Models

Llama 4 Maverick Debuts with 128 Expert Mixture Architecture

Meta releases Llama 4 Maverick, featuring a massive 128-expert mixture-of-experts architecture with 400B total parameters but only 17B active per token. The model matches frontier closed-source models while remaining open weight.

meta llama open-source +2

Feb 04, 2026 · Meta

Read

Video Models

Meta Movie Gen Open Sources Video Generation Pipeline

Meta releases Movie Gen as open source, providing the complete video generation pipeline including model weights, training code, and a comprehensive research paper. The release democratizes high-quality AI video generation.

meta movie-gen open-source +2

Feb 04, 2026 · Meta

Read

AI Agents & Tools

CrewAI Raises $50M Series B for Multi-Agent Orchestration

CrewAI secures $50M in Series B funding to scale its multi-agent orchestration platform. The funding will accelerate enterprise adoption and development of advanced agent collaboration features including hierarchical task delegation.

crewai multi-agent funding +2

Feb 03, 2026 · CrewAI

Read

Image Models

Google Imagen 4 Sets New FID Record on ImageNet

Google DeepMind releases Imagen 4 with a record-breaking FID score of 1.42 on ImageNet, demonstrating unprecedented image quality. The model introduces native multi-language prompting and cultural context awareness.

google imagen fid +2

Feb 03, 2026 · Google

Read

Open Source AI

Together AI Platform Reaches 200+ Open Source Models

Together AI announces that its inference platform now hosts over 200 open-source models with optimized serving, automatic scaling, and competitive pricing. The milestone reflects the rapid expansion of the open-source AI ecosystem.

together-ai inference open-source +2

Feb 02, 2026 · Together AI

Read