Gemini 3.5 Pro X-High, MiniMax M3, DeepSwe, New Claude Models, MiMO-v2.5 Upgrade, & More! AI NEWS

· Source: WorldofAI · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Intermediate, long

Summary

Google is reportedly preparing to launch Gemini 3.5 Pro X-High, an "extra high thinking variant" to improve reasoning in long-horizon tasks, potentially in June. Additionally, a Gemini Live model (Gemini 3.1 Flash Live VR EAP) with voice cloning for real-time multimodal interaction is anticipated. Miniax teased its M3 model, featuring a sparse attention architecture that could achieve 10x faster context processing and 15x faster decoding, enabling ultra-long context AI with lower compute. Anthropic's Claude Lab products, including "claude spaces," suggest an expansion into collaborative workspaces and persistent agent environments. Xiaomi's MiMO 2.5 Pro now offers significantly reduced API costs (up to 99%) and increased tokens, matching Deepseek v4 Pro's pricing. New benchmarks include Deep Sway for agentic coding, where OpenAI's GPT 5.5 scored ~70%, and Quen 3.7 Max, ranking #4 on Code Arena. Figure AI is commercially deploying humanoid robots with Catalyst Brands, starting in Reno, Nevada.

Key takeaway

For Machine Learning Engineers evaluating model architectures, investigate sparse attention techniques like Miniax's M3 for ultra-long context efficiency and reduced compute. If you are a Director of AI/ML managing API costs, re-evaluate providers like Xiaomi's MiMO 2.5 Pro, which now offers competitive pricing and increased tokens. Teams developing agentic systems should explore new benchmarks like Deep Sway to accurately assess model performance on realistic software engineering tasks, informing model selection for complex workflows.

Key insights

Major AI model updates, architectural innovations, and commercial deployments are rapidly advancing AI capabilities and applications.

Principles

Method

Miniax's sparse attention performs a lightweight scan to identify relevant sections, then focuses heavy reasoning only on those areas, similar to using a textbook index.

In practice

Topics

Best for: CTO, VP of Engineering/Data, AI Architect, AI Scientist, Machine Learning Engineer, Director of AI/ML

Related on AIssential

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by WorldofAI.