Hub de Notícias de IA

Destaque

Video Models

Sora 2 Now Available to All ChatGPT Plus Subscribers

OpenAI rolls out Sora 2 to all ChatGPT Plus subscribers, enabling 1080p video generation up to 60 seconds. The updated model features improved temporal consistency, physics understanding, and audio synchronization.

openai sora video-generation +2

Feb 24, 2026 · OpenAI

Ler

Destaque

AI Agents & Tools

Anthropic MCP Becomes the Industry Standard for AI Tool Integration

Anthropic's Model Context Protocol (MCP) has been adopted by major AI platforms as the standard for connecting AI models to external tools and data sources. Over 500 MCP servers are now available in the open-source ecosystem.

anthropic mcp tool-integration +2

Feb 23, 2026 · Anthropic

Ler

Destaque

AI Coding Agents

Claude Code Launches Parallel Agent Architecture

Anthropic introduces a groundbreaking parallel agent architecture for Claude Code, enabling multiple sub-agents to work on different parts of a codebase simultaneously. The new system reduces complex refactoring tasks from hours to minutes.

claude coding agents +2

Feb 23, 2026 · Anthropic

Ler

Destaque

Open Source AI

Llama 4 Scout and Maverick Set New Open Source Benchmarks

Meta releases the complete Llama 4 family including Scout (17B active MoE) and Maverick (128-expert MoE), establishing new performance records for open-weight models across reasoning, coding, and multilingual benchmarks.

meta llama-4 open-source +2

Feb 22, 2026 · Meta

Ler

Destaque

Text Models

Claude Opus 4.6 Achieves New State-of-the-Art Benchmarks

Anthropic releases Claude Opus 4.6, the most capable model in the Claude family, achieving state-of-the-art results across reasoning, coding, and mathematical benchmarks. The model introduces enhanced extended thinking capabilities.

claude opus anthropic +2

Feb 22, 2026 · Anthropic

Ler

Destaque

Image Models

GPT Image 1.5 Achieves Near-Photorealistic Generation Quality

OpenAI announces GPT Image 1.5 with dramatically improved photorealism, text rendering accuracy, and compositional understanding. The model generates images nearly indistinguishable from photographs in blind evaluation tests.

openai image-generation photorealism +2

Feb 21, 2026 · OpenAI

Ler

AI Coding Agents

GitHub Copilot X Integrates GPT-5 for Full Codebase Understanding

GitHub announces Copilot X powered by GPT-5, delivering full repository-level code understanding with semantic search across entire codebases. The upgrade brings 10x context window improvements and multi-file editing capabilities.

github copilot gpt-5 +2

Feb 20, 2026 · GitHub

Ler

Video Models

Google Veo 3 Generates 4K 60fps Video with Audio

Google DeepMind unveils Veo 3, capable of generating 4K resolution video at 60fps with synchronized audio. The model introduces storyboard-based control and multi-scene narrative generation for long-form content.

google veo video-generation +2

Feb 20, 2026 · Google

Ler

Text Models

GPT-5.2 Introduces Extended Thinking with Chain-of-Thought

OpenAI releases GPT-5.2 with a new extended thinking mode that combines traditional chain-of-thought prompting with learned reasoning strategies. The model shows a 15% improvement on complex reasoning tasks over GPT-5.

openai gpt-5 chain-of-thought +2

Feb 19, 2026 · OpenAI

Ler

AI Agents & Tools

OpenAI Operator Enables Multi-Step Browser Automation

OpenAI launches Operator, an AI agent that can autonomously perform multi-step tasks in a web browser. The agent can book flights, fill out forms, and manage online accounts with human-in-the-loop confirmation for sensitive actions.

openai operator browser-automation +2

Feb 19, 2026 · OpenAI

Ler

Image Models

Midjourney v7 Introduces 3D Scene Generation and Spatial Control

Midjourney launches version 7 with revolutionary 3D scene generation capabilities, allowing users to create explorable 3D environments from text prompts. The update also includes spatial composition controls and multi-view consistency.

midjourney 3d image-generation +2

Feb 18, 2026 · Midjourney

Ler

Open Source AI

DeepSeek R2 Achieves GPT-4 Level at 5% of the Cost

DeepSeek releases R2, a reasoning-focused open-source model that matches GPT-4 Turbo on key benchmarks while costing 95% less to serve. The model uses a novel reinforcement learning approach for improved mathematical and logical reasoning.

deepseek r2 open-source +2

Feb 18, 2026 · DeepSeek

Ler