English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

Easy AI Daily News Roundup - June 11, 2025

Forum topic · 小凯 · 2026-03-27

Summary

A roundup of AI industry news for June 11, 2025. Meta acquired 49% of Scale AI for $15 billion and hired founder Alexandr Wang to lead its new Superintelligence lab. OpenAI released o3-pro for ChatGPT Pro and API users, cut o3 prices by 80% ($2/$8 per million tokens), and enabled DPO fine-tuning for GPT-4.1 models. Volcano Engine launched Doubao 1.6 and the Seedance 1.0 Pro video generation model, while Krea AI released its Krea1 image model. Alibaba open-sourced the MaskSearch pre-training framework, Mistral AI released the 24B Magistral reasoning model under Apache 2.0, and Meta introduced the V-JEPA 2 world model for zero-shot robot planning. Gemini 2.5 Pro topped live benchmarks, OpenAI reportedly signed a cloud deal with Google, Modular added AMD Instinct GPU support, Figma launched an MCP server beta, and Disney and Universal sued Midjourney over copyright infringement.

Personnel Moves

  • Meta builds Superintelligence lab: Meta acquired 49% of Scale AI for $15 billion and hired founder Alexandr Wang to lead its new "Superintelligence" division, reportedly offering top researchers eight-figure salaries. Jack Rae joined from Gemini. Details: The Verge
  • Cross-industry executive moves: Lamini founder Sharon Zhou joined AMD as VP of AI, reporting to Lisa Su. Former Instacart CEO Fidji Simo became CEO of OpenAI's Applications division, reporting to Sam Altman. Source
  • Model Updates and Pricing

  • OpenAI o3-pro released: Available to ChatGPT Pro users and via API. Scores 87.3 on Extended NYT Connections (vs. 82.5 for o3), tops SnakeBench, and solves 10-disc Tower of Hanoi. Users report it "far exceeds Claude Opus 4" on non-code tasks, though inference is 3x slower than o1-pro. Source
  • OpenAI cuts o3 pricing by 80%: Now $2 per million input tokens and $8 per million output tokens — 20% cheaper than GPT-4o — potentially forcing Google and Anthropic to follow. Source
  • DPO fine-tuning for GPT-4.1: GPT-4.1, 4.1-mini, and 4.1-nano now support Direct Preference Optimization for improving style and creativity on subjective tasks. Source
  • New Models and Research

  • Volcano Engine launches Doubao 1.6 and Seedance 1.0 Pro: Announced at the FORCE conference. Doubao 1.6 uses unified pricing, supports "search-while-thinking" and DeepResearch. Seedance 1.0 Pro leads in text-to-video and image-to-video generation. More info
  • Krea AI releases Krea1: First image generation model from Krea, addressing texture, contrast, and composition issues; supports style references and custom training, with a free beta. Demo
  • Alibaba open-sources MaskSearch: From the Tongyi Lab, using Retrieval-Augmented Mask Prediction (RAMP) and reinforcement learning to improve complex question answering. Code: https://github.com/Alibaba-NLP/MaskSearch
  • Mistral releases Magistral: A 24B-parameter reasoning model based on Mistral Small 3.1, multilingual with 128K context (40K effective), under Apache 2.0. Model card
  • Meta V-JEPA 2: A 1.2B-parameter video-trained world model enabling zero-shot robot planning in unseen environments, plus three new physical-reasoning benchmarks. Blog
  • Gemini 2.5 Pro gains: The 06-05 checkpoint tops Live Fiction (192K tokens) and document processing, reportedly solving all 2025 JEE Advanced math problems. Source
  • Partnerships and Infrastructure

  • OpenAI–Google cloud deal (reported): OpenAI reportedly signed a new cloud agreement with Google for additional compute. Source
  • Modular supports AMD GPUs, launches Mammoth: Added AMD Instinct MI300X/MI325 support with 53% BF16 throughput gains, plus Mammoth, a Kubernetes-native system for scaling GenAI inference across GPU clusters. Details
  • Figma Dev Mode MCP server beta: Passes design data to AI coding tools via a standardized protocol, improving code output fidelity from prototype to production. Details
  • Legal and Ethics

  • Disney and Universal sue Midjourney: The studios allege unauthorized use of copyrighted characters (e.g., Star Wars, The Simpsons) for training — a case that could shape AI training-data copyright standards. Reuters
---

*Source: Easy AI Daily*

Tags

#ai-news#openai#meta#mistral#alibaba#world-models#copyright#daily-roundup

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177169119