AI News Roundup: Google Launches Gemma 4, OpenAI Shares Go Cold, Alibaba Drops Qwen3.6-Plus

Google Launches Gemma 4 — Its Most Capable Open Models Yet

Google released Gemma 4 on April 2, a new family of open models purpose-built for advanced reasoning and agentic workflows. Available in four sizes from 2 billion to 31 billion parameters, Gemma 4 is built on technology from Google’s proprietary Gemini 3 models and released under the commercially permissive Apache 2.0 license.

The models are multimodal, run completely offline with near-zero latency, and target edge devices including phones, Raspberry Pi, and NVIDIA Jetson Orin Nano. Google says Gemma 4 delivers “an unprecedented level of intelligence-per-parameter,” and the models will serve as the foundation for the next generation of Gemini Nano — meaning code written for Gemma 4 today will automatically work on Gemini Nano 4-enabled devices later this year.

Day-one tool support is extensive: Hugging Face, vLLM, llama.cpp, MLX, Ollama, NVIDIA NIM, LM Studio, and more. NVIDIA simultaneously announced accelerated support for Gemma 4 across its RTX and Jetson platforms, signaling strong industry buy-in for Google’s open model strategy.

$600 Million in OpenAI Shares Can’t Find a Single Buyer

OpenAI shares have fallen sharply out of favor on the secondary market. According to Bloomberg, about half a dozen institutional investors — including hedge funds and venture capital firms holding large stakes — approached secondary marketplaces looking to sell roughly $600 million in OpenAI shares. Not a single buyer emerged.

The contrast with rival Anthropic is stark: buyers across multiple secondary platforms reportedly have $2 billion in cash ready for Anthropic shares. The investor rotation is driven by Anthropic’s lower valuation ($380 billion vs. OpenAI’s $852 billion), cleaner governance structure, and increasingly competitive model performance. OpenAI’s annualized revenue now tops $25 billion, but Anthropic is closing fast at nearly $19 billion.

The shift is particularly notable given OpenAI just closed a $122 billion funding round — the largest in venture history — and is reportedly taking early steps toward a public listing. Whether the secondary market freeze is a blip or a structural repricing remains one of the biggest questions in AI investing right now.

Alibaba Unveils Qwen3.6-Plus for Agentic AI

Alibaba released Qwen3.6-Plus on April 2, its third proprietary AI model in just a few days, signaling an aggressive push to monetize its AI capabilities. The model supports a 1-million-token context window by default and is designed for agentic coding, multimodal perception, and autonomous reasoning across complex engineering tasks.

Qwen3.6-Plus can plan, test, and iterate on code at the repository level, analyze images, documents, and videos, and deliver high-accuracy instruction following for demanding real-world scenarios like retail intelligence and automated inspections. The model is being integrated into Alibaba’s enterprise platform Wukong and is compatible with third-party coding tools including OpenClaw, Claude Code, and Cline.

States Move Fast on AI Chatbot Safety Laws

AI chatbot regulation is accelerating at the state level. Tennessee Governor Bill Lee signed SB 1580, which prohibits the deployment of any AI system that represents itself as a qualified mental health professional. The bill passed the Senate 32–0 and the House 94–0 — a rare unanimous consensus.

In Georgia, three AI-related bills are on Governor Kemp’s desk before the April 6 adjournment: SB 540, a chatbot disclosure and child safety bill requiring AI systems to identify themselves to minors and include protocols for suicidal ideation; SR 789, creating a Senate AI study committee; and SB 444, which prohibits insurance coverage decisions from being based solely on AI. Across the country, 78 AI chatbot safety bills have been introduced in 27 states this legislative session.

Crisis Contractor ThroughLine Expands from Self-Harm to Extremism

ThroughLine, the New Zealand startup hired by OpenAI, Anthropic, and Google to redirect users to crisis support when flagged for self-harm, domestic violence, or eating disorders, is developing a new tool to combat violent extremism. The company is in discussions with The Christchurch Call, the anti-extremism initiative formed after New Zealand’s 2019 terrorist attack.

The initiative comes as AI companies face mounting lawsuits over safety failures. OpenAI was threatened with intervention by the Canadian government in February after revealing a person who carried out a deadly school shooting had previously been banned by the platform without authorities being notified. No timeline has been set for rollout, and questions remain about whether flagging dangerous users could itself trigger escalated behavior.