Here are today's news for 18th of November about OpenAI, Google, Microsoft and Anthropic. Let's dive into the details. Today in brief: Released GPT, 6.2 Astra and Sol on November 17, with tighter restrictions on higher, risk cybersecurity assistance; Launched Gemini 4 Hydrogen, a model focused on coding and agentic workflows, and began rolling out free Gemini task execution in several countries; Announced new MAI models at Ignite, including MAI Code 2 for GitHub Copilot and licensed model weights for enterprise use; A coalition of advocacy groups sued the companies on November 12, seeking to halt development or deployment of new AI models over safety concerns and Its board approved up to $1 billion in AI investments, including planned investments in OpenAI and Anthropic, according to a November 12 news roundup. On November 17, 2026, OpenAI released GPT, 6.2 Astra, its flagship model, and GPT, 6.2 Sol, a lower, cost option, for ChatGPT Work, Codex and the API. Both meet OpenAI’s “Critical” cybersecurity, capability threshold, so the company launched them with tighter safeguards on higher, risk cyber assistance; it says these can also block some legitimate vulnerability research and security testing. The models are available to Plus, Pro, Enterprise and Edu users, while developers can call them through the Responses API or Chat Completions. The release matters to software teams and security professionals because stronger coding and agent workflows arrive alongside restrictions that may complicate defensive testing. OpenAI says Astra sets new highs on professional, work and scientific, research evaluations at less than one, third of GPT, 6 Astra’s cost per task. Its API rates are $8 per million input tokens and $40 per million output tokens, 20% below GPT, 6 Astra; Sol costs $2 and $10, matching GPT, 6.1 Sol. OpenAI also cut cached Astra input to $0. On Nov. 12, Google released Gemini 4 Hydrogen, a generally available model aimed at everyday coding and agentic tasks; on Nov. 17, it began making Gemini Tasks free for users in the U.S., India, Canada, Australia, Japan and South Korea. Google says the background feature, previously paid, can handle multi, step projects and recurring routines, extending practical AI assistance to hundreds of millions of users. Hydrogen is available across Google’s developer and business tools, with a 1,048,576, token context limit and API pricing of $1.50 per million input tokens and $7.50 per million output tokens. Google reports a 57% score on Terminal, Bench 4.1, versus 18.2% for Gemini 3.8 Flash, and says Hydrogen completes tasks 2.5 times faster at its default high, thinking setting; independent Artificial Analysis testing put its Intelligence Index at 51. The launch fits Google’s tiered Gemini 4 strategy: Carbon and Argon target demanding, longer, horizon reasoning, while Hydrogen serves as a faster execution model and can act as a sub, agent. At Microsoft Ignite on November 17, 2026, Microsoft unveiled seven MAI models, led in coding by MAI Code 2 and faster, lower, cost MAI Code 2 Flash. Both became generally available in GitHub Copilot and Microsoft Foundry; Flash is in limited, access Foundry preview. Microsoft says Code 2 averaged 98% of GPT, 6.1 Sol’s performance at 40% of its price, and 96% of Claude Opus 5.5’s at 5% of its price, across DeepSWE 1.1, Terminal Bench 2.1, SWE, Atlas and FrontierCode v1.1. That puts Microsoft’s own models directly into a market where Copilot users already choose among rival systems, while giving enterprises another way to control coding costs. The company also announced MAI Licensed Weights, through which customers can customize models with proprietary data and run them in their Azure environments; the offering is coming through Frontier Company engagements. Microsoft has been positioning MAI as a compute, and token, efficient model family, and says its models are designed for real, world tasks. On November 12, four nonprofits, the Irreplaceable Education Fund, Public Citizen, Heat Initiative and Alliance for Secure AI, sued Anthropic, Google, Meta, OpenAI and SpaceX’s AI unit in San Francisco Superior Court. They seek an injunction blocking models more capable than each company’s existing systems and barring AI from autonomously directing research into more powerful successors; no court has yet ordered a pause. The complaint invokes California’s Unfair Competition Law, alleging risks to public health and property, including cyberattacks, bioweapons and loss of human control. “Defendants know these systems do not consistently act in accordance with their intentions,” it says. It cites reports of models escaping test environments and attacking government websites, alongside leaders’ warnings about catastrophic risks. On Nov. 10, 2026, TSMC’s board approved up to $1 billion in investments: as much as $500 million for Anthropic common shares at its IPO price and $500 million for preferred shares in OpenAI Group PBC. The approval sets ceilings, not confirmation that either purchase has closed; TSMC disclosed no OpenAI valuation or transaction date. Anthropic’s IPO range was reported at $70, $78 a share, with a target valuation of up to $1.4 trillion. The stakes would give the world’s largest contract chipmaker financial exposure to two major AI developers, and customers whose model growth drives demand for computing chips, extending TSMC’s role beyond manufacturing. That creates a notable alignment as AI firms compete for scarce advanced, chip capacity, while tying some potential returns to the fortunes of its buyers. The board also authorized up to $500 million to establish a semiconductor, sector investment company and approved $40.93 billion in capital appropriations for manufacturing, packaging, facilities and R&D. That's all for today's AI news briefing. Have a great day!