Google has launched Gemini 4 Argon, positioning it for complex workflows, software engineering, knowledge work, and cybersecurity defense. It is currently being made available to trusted cybersecurity personnel through the Fairwind program, with plans for broader access yet to be determined. The single-output limit has been increased to 1 million tokens. The official launch pricing is listed at $2 per million input tokens and $10 per million output tokens. All positioning, availability, and performance information is per Google’s official release.<\/p>\n
Sources:<\/p>\n
The White House has announced the “White House Accord on Super Intelligence.” NVIDIA CEO Jensen Huang summarized it as placing safety responsibility on the companies developing and deploying the technology, emphasizing internal controls, independent external evaluation, and board oversight. MaxForAI’s analysis of the accord text notes it is a voluntary agreement, not current law; thus, it reflects the governance commitments of the signatories and should not be equated with enacted regulatory requirements.<\/p>\n
Sources:<\/p>\n
OpenAI executive Tibo stated that GPT-6.1 Sol has become one of the company’s most in-demand models on both the API and subscription sides, with ChatGPT and Codex experiencing significant load. Following the deployment of additional capacity, service speed is expected to be nearly double that of the previous day. Sam Altman later called it OpenAI’s fastest-growing model. The demand and performance descriptions come from company executives, reflecting the initial supply-demand dynamics and scaling efforts post-launch.<\/p>\n
Sources:<\/p>\n
OpenAI’s developer account introduced that ChatGPT’s Sites and plugins can now be centrally displayed on shareable personal pages, making them easier for others to discover and reuse. Tibo added that users can now directly build and deploy MCP servers within ChatGPT, restricting access or sharing them publicly. This extends ChatGPT from a single conversational interface into a customizable, distributable small-scale tool platform.<\/p>\n
Sources:<\/p>\n
OpenClaw officially released v2026.9.7, listing improvements such as faster responses under high load, smoother long conversations, backup and rollback updates, fixes for the 9.5 upgrade, and restart recovery. It also adds OpenAI Agents API and ChatGPT login Beta. The release post notes this version includes 2,818 PRs from 344 contributors, indicating a broad-scope update.<\/p>\n
Sources:<\/p>\n
Figure officially released a project detailing the retirement of F.02 by entering a 75-ton electric arc furnace in Finland. A summary by Dotey states that the team trained an action model in a simulation environment, enabling the robot to perform the jump into the molten steel container at a foundry it had not previously surveyed. On-site filming and equipment conditions imposed constraints different from typical lab demos. Details of the autonomous action are from the blogger’s account, while the official post confirms the project and furnace scenario.<\/p>\n
Sources:<\/p>\n
Flow founder Pari Singh announced the company has completed a $50 million Series B funding round at a $750 million valuation. The introduction positions Flow as a platform for hardware teams to manage requirements and verification processes, suggesting AI is evolving from integrating CAD, analysis, and simulation work towards directly taking on some execution tasks. This is the company’s description of its own business and industry direction.<\/p>\n
Sources:<\/p>\n