ChatGPT Work Launch, Open-Weight Debate, and New Model Releases Dominate AI News

Key Takeaways

1. OpenAI released ChatGPT Work, providing persistent agentic workflows from mobile, including cloud computers, terminals, and automated tasks.
2. The open-weight model debate intensified, with Hugging Face, Yann LeCun, and Demis Hassabis advocating for openness, while Elon Musk announced X system code will be open-sourced.
3. Claude Opus 5 achieved 30.2% SOTA on ARC-AGI-3, Midjourney released V8.2, and Grok 4.5 demonstrated strong performance, highlighting rapid model competition.
4. NVIDIA announced major AI infrastructure investments in South Korea, including a 2GW AI factory and next-gen memory collaboration.

1. ChatGPT Work Launches with Advanced Agent Capabilities

1. Sam Altman highlighted ChatGPT Work's ability to access cloud computers, persistent workspaces, terminals, remote browsers, and integrated tools like Git, scheduling, and website building — all from a phone, eliminating the need for a laptop. — via 1
2. Greg Brockman added that the agent feature now supports persistent login sessions across websites, enabling complex automation, and emphasized that agentic abilities are core to AI's impact. — via 1 2

2. Open-Weight Model Debate Heats Up

1. Hugging Face argued that open-weight models are essential for a healthy AI ecosystem, citing a concrete event where the first autonomous AI attack was launched by a closed-source model and defended by an open-source model. — via 1 2
2. Yann LeCun strongly criticized the push to restrict open-weight models for safety, alleging it is a monopolistic move by companies like Anthropic and OpenAI, and emphasized that open models foster competition. — via 1
3. Demis Hassabis stressed Google's commitment to openness, noting Gemma models have surpassed 300 million downloads, and proposed a framework for responsible deployment of both open and proprietary models. — via 1
4. Aravind Srinivas argued that both frontier closed and open models are necessary, and that open models provide a security advantage by allowing more teams to inspect and harden them. — via 1 2

3. New Model Releases and Performance Highlights

1. Claude Opus 5 is now available on Perplexity, achieving 30.2% SOTA on ARC-AGI-3 (up from 7.8%) and outperforming several models on WANDR while being 57% cheaper. Ethan Mollick reported it fully replaces Opus 4.8 but noted language quirks (density and "FableSpeak"). — via 1 2 3
2. Midjourney released V8.2 as the default model, with major improvements in aesthetics, personalization, and image quality, offering more creative and bold styles. — via 1
3. Elon Musk announced Grok Build updates including a guided tutorial, smart search rewrite, and workflow progress display. Grok 4.5 achieved the highest perfect extraction rate on invoice processing and surpassed GPT-5.6 Sol and Fable 5 on OpenRouter token traffic. — via 1 2 3 4
4. NVIDIA's optimization reduced DeepSeek-V4 Pro startup time from 8 minutes to under 2 minutes using GPU RDMA and ModelExpress, a technique that also accelerates inference and RL post-training. — via 1

4. AI Infrastructure and Chip Competition

1. NVIDIA announced collaborations in South Korea: SK Telecom will build a 2GW NVIDIA Vera Rubin DSX AI factory, and SK Hynix will co-develop next-gen AI memory including HBM. NAVER also partners with NVIDIA and Brookfield to expand its national AI factory. — via 1 2
2. Deedy Das noted multiple next-gen AI chip startups are targeting specific bottlenecks (DRAM, interconnect, compute/memory separation) to challenge Nvidia's dominance, with distinct technological approaches. — via 1