China’s New Self Improving Open AI Beats OpenAI

Author: AI Revolution - Bewertung: 20x - Views:227

MiniMax just open-sourced M2.7, a new self-improving AI model built for coding, software engineering, office work, and multi-agent automation. Google is adding voice control to Mixboard, OpenAI is developing a unified Codex agent app, Meta just launched Muse Spark with parallel reasoning, and Runable just introduced RunClaw for delegated AI work inside team chat tools.



📩 Brand Deals & Partnerships: [email protected]

✉ General Inquiries: [email protected]

New Channel: https://www.youtube.com/@science.revolution



🧠 What You’ll See

MiniMax M2.7 open-source release and benchmarks

SOURCE: https://huggingface.co/MiniMaxAI/MiniMax-M2.7



Runable RunClaw AI agent launch and $2M ARR milestone

SOURCE: https://runable.com/runclaw



Google Mixboard voice mode and workspace features

SOURCE: https://labs.google.com/mixboard/welcome



OpenAI unified Codex app and Scratchpad

SOURCE: https://www.testingcatalog.com/openai-develops-unified-codex-app-and-new-scratchpad-feature/



Meta Muse Spark and Contemplating mode

SOURCE: https://ai.meta.com/blog/introducing-muse-spark-msl/



Timestamps:

0:00 Intro

0:18 MiniMax M2.7 Open-Sourced

1:16 Engineering & Office Benchmarks

3:11 Self-Evolving AI Breakthrough

6:27 Runable RunClaw AI agent

8:53 Google Mixboard Goes Voice

9:44 PDF Export & Collaboration

10:14 OpenAI Codex Unified App

10:22 Scratchpad & Managed Agents

11:39 Meta Muse Spark Launch



🚨 Why It Matters

AI is moving from prompt tools into systems that can plan, reason, and execute real work with less human input. MiniMax is pushing self-improving open models, Google is testing voice-driven collaboration, OpenAI is moving toward managed agent workflows, Meta is scaling multi-agent reasoning, and Runable is bringing delegated AI work directly into team chats.



#ai #artificialintelligence #openai