Rakuten releases RakutenAI-3.0, 671B-parameter Japanese-optimized mixture-of-experts model

NVIDIA released Nemotron-3-Nano-4B-GGUF on March 16, 2026, a 4-billion parameter small language model (SLM) designed for edge deployment on devices like Jetson Thor and GeForce RTX. The model features a hybrid Mamba-2 and Transformer architecture with a 262K token context window and supports both reasoning and non-reasoning modes via system prompts.

Nvidia has released Nemotron 3 Super, a 120-billion parameter hybrid Mamba-Transformer Mixture-of-Experts model that activates only 12 billion parameters during inference. The open-weight model features a 1-million token context window, multi-token prediction capabilities, and pricing at $0.10 per million input tokens and $0.50 per million output tokens.

Xiaomi released MiMo-V2-Pro on March 18, 2026, a flagship foundation model with over 1 trillion total parameters and a 1,048,576 token context window. The model is priced at $1 per million input tokens and $3 per million output tokens, positioning it as an agent-focused system comparable to top-tier models.

Multiverse Computing, a Spanish startup, has launched a self-serve API portal giving developers direct access to compressed versions of models from OpenAI, Meta, DeepSeek, and Mistral AI. The move targets enterprises seeking to reduce cloud infrastructure dependence and lower compute costs through edge deployment. The company claims its HyperNova 60B 2602 model delivers faster responses at lower cost than the original OpenAI model it was derived from.

Adobe is rolling out custom models for Firefly, allowing creators to train the generative model on 10-30 of their own images to generate new content matching their specific visual style. The feature costs 500 credits per training session and supports three methods: photography style, illustration style, and character consistency.

NVIDIA Nemotron 3 Super, a hybrid Mixture of Experts model with 120B parameters and 12B active parameters, is now available as a fully managed model on Amazon Bedrock. The model supports up to 256K token context length and claims 5x higher throughput efficiency over the previous Nemotron Super and 2x higher accuracy on reasoning tasks.

Google has launched a vibe coding feature in Google AI Studio that converts natural language descriptions into working applications using Gemini 3.1 Pro. The platform now supports real-time multiplayer games and automatically configures databases, authentication, and third-party service integrations through an "Antigravity Agent."

Microsoft's superintelligence team, led by Mustafa Suleyman, has released MAI-Image-2, a text-to-image generator that currently ranks third on the Arena.ai leaderboard for text-to-image models, behind OpenAI's GPT-Image-1.5 and Google's Nano Banana 2. The model is now available for testing in the MAI Playground and will roll out to Copilot and Bing Image Creator, with API access opening to all developers through Microsoft Foundry.

Anthropic has added "channels" to Claude Code, enabling Claude to respond to incoming messages, webhooks, and notifications asynchronously without user intervention. The research preview supports Telegram and Discord with custom channel support, running through MCP servers with two-way communication.

NVIDIA released Nemotron 3 Content Safety 4B, an open-source multimodal safety model designed to moderate content across text, images, and multiple languages. Built on Gemma-3 4B-IT with a 128K context window, the model achieved 84% average accuracy on multimodal safety benchmarks and supports over 140 languages through culturally-aware training data.

OpenAI has acquired Astral, the company behind widely-used Python development tools Ruff, uv, and ty, which see hundreds of millions of downloads monthly. The tools will integrate with OpenAI's Codex platform to enable agentic AI systems that can plan changes, modify codebases, run tools, and verify results. OpenAI commits to keeping the tools open source under permissive licenses post-acquisition.

Google has launched task automation for Gemini on Pixel 10 Pro and Galaxy S26 Ultra, allowing the AI to autonomously use apps for food delivery and rideshare services. The feature works but is slow—taking approximately nine minutes to complete an order—and remains limited to a small beta subset of apps. Despite performance limitations, it represents the first practical demonstration of an AI assistant actually controlling a phone outside of controlled demos.