LLM News

IBM released Granite 4.1, a family of dense decoder-only LLMs (3B, 8B, 30B parameters) trained on approximately 15 trillion tokens using a five-phase pre-training pipeline. The 8B instruct model matches or surpasses the previous Granite 4.0-H-Small (32B-A9B MoE) despite using fewer parameters and a simpler dense architecture. All models support up to 512K context windows and are released under Apache 2.0 license.

Mistral AI has integrated its Medium 3.5 model into Vibe for remote coding agent functionality. The company also launched a new Work mode in Le Chat designed for complex tasks, though specific technical details remain undisclosed.

AWS has released support for custom Model Context Protocol (MCP) proxies on Amazon Bedrock AgentCore Runtime, allowing organizations to implement custom governance and security controls on AI agent tool interactions without modifying upstream MCP servers. The serverless proxy runs on AgentCore Runtime with automatic scaling and built-in observability through CloudWatch and OpenTelemetry.

OpenAI released GPT-5.5 on April 23, its first retrained base model since GPT-4.5, scoring 82.7% on Terminal-Bench 2.0 versus GPT-5.4's 75.1% and Claude Opus 4.7's 69.4%. API pricing is set at $5 per million input tokens and $30 per million output tokens, exactly double GPT-5.4 rates.

NVIDIA has released Nemotron 3 Nano Omni 30B-A3B, a multimodal large language model with 31 billion parameters using a Mamba2-Transformer hybrid Mixture of Experts architecture. The model supports video, audio, image, and text inputs with a 256K token context window and includes a dedicated reasoning mode with chain-of-thought capabilities.

Poolside has released Laguna XS.2, a 33B total parameter Mixture-of-Experts model with 3B activated parameters per token, designed for agentic coding. The model features a 131,072-token context window, scores 68.2% on SWE-bench Verified, and is available under Apache 2.0 license with free API access.

Lovable, an AI-assisted programming platform, launched an iOS app that generates web applications from natural language and voice prompts. The app allows users to start projects on mobile and continue on desktop, with subscription plans ranging from $9.99 to $79.99 per month.

Meta released Muse Spark in early April, its first closed-source AI model that will eventually offer paid developer access. According to Arena.AI rankings, Muse Spark trails Anthropic's Claude and Google's Gemini in text capabilities but beats OpenAI's GPT in vision tasks.

AWS has added OpenAI's latest models, Codex, and a new managed agents service to its Bedrock platform, one day after OpenAI revised its agreement with Microsoft. The integration follows OpenAI's up-to-$50 billion deal with Amazon.

Amazon Web Services released technical guidance for migrating text agents to voice assistants using Amazon Nova 2 Sonic, a native speech-to-speech model that combines automatic speech recognition, reasoning, tool calling, and text-to-speech in a single bidirectional streaming interface. The model supports asynchronous tool calling and built-in voice activity detection for handling interruptions.

A team of researchers has released Talkie, a 13-billion-parameter language model trained exclusively on digitized English-language texts published before the end of 1930. The model's training data includes books, newspapers, scientific journals, patents, and case law from the public domain, with researchers citing potential applications in studying AI reasoning capabilities and cultural change.

IBM has launched Bob, its AI coding assistant, following internal testing with 80,000 employees. The company claims teams saw average productivity gains of 45% across complex workflows. Pricing ranges from $20 to $200 per month using a "Bobcoin" credit system.

Amazon has released a desktop version of its Quick AI assistant that integrates with Google Workspace, Microsoft 365, Zoom, and Salesforce, storing persistent context about user activities to automate tasks. The company also split Amazon Connect into four vertical-specific products: Connect Decisions, Connect Talent, Connect Health, and Connect Customer AI.

Google's Gemini for Home voice assistant now executes smart home commands up to 1.5 seconds faster for lights and plugs, the company announced. The update also brings near-instant processing for alarms, timers, and reminders, currently available for English, French, and Spanish users.

Anthropic has released connectors that enable Claude to integrate directly with creative software including Adobe Creative Cloud, Blender, Ableton, Autodesk, and Affinity. The company also became a Corporate Patron of the Blender Development Fund, committing at least €240,000 ($281,000) annually to support the open-source 3D software.

NVIDIA has launched Nemotron 3 Nano Omni on Amazon SageMaker JumpStart, a multimodal model with 30 billion total parameters (3 billion active) that processes video, audio, images, and text in a single inference pass. The model features a 131K token context window and uses a Mamba2 Transformer Hybrid MoE architecture combining three specialized encoders.

Nvidia has released Nemotron 3 Nano Omni, a 30-billion-parameter multimodal model available free on OpenRouter. The model features a 256,000-token context window, accepts text, image, video, and audio inputs, and claims 2× higher throughput for video reasoning compared to separate vision and speech pipelines.

Perplexity rolled out native iPadOS support for its Comet AI browser on April 28, 2026. The update adds multiple window support and Split View multitasking, allowing users to run Comet alongside other iPad apps.

Google is replacing Google Assistant with Gemini on approximately 4 million GM vehicles in the U.S. running Android Automotive. The rollout begins with model year 2022 and newer Buick, Cadillac, Chevrolet, and GMC vehicles.

Lovable has launched its vibe-coding app on iOS and Android app stores, allowing users to build web apps through voice or text prompts on mobile devices. The launch comes after Apple blocked updates to competitors like Replit and Vibecode, forcing vibe-coding apps to preview generated code in web browsers rather than within the app itself.