// NODE-02 · NEURAL FEED · DAILY TRANSMISSION //
June 28, 2026 · WINDYVIEWS.COM
Google has pushed the public launch of Gemini 3.5 Pro to July 2026. Announced at I/O on May 19, only the faster Gemini 3.5 Flash shipped that day; Pro remains in limited Vertex AI enterprise preview. The frontier model targets a 2-million-token context window and a "Deep Think" reasoning mode. The slip coincides with a talent drain, as several senior Google researchers have left for Anthropic.
On June 9, Anthropic launched Claude Fable 5 and Claude Mythos 5, a capability tier above Claude Opus 4.8 — which itself sits at #1 on the Artificial Analysis Intelligence Index with a 61.4 score, the first model cleanly above 60. Three days later, the US government issued an urgent export-control directive citing national security, barring access to the two new models by any foreign national.
Sensor Tower's 2026 State of AI report shows ChatGPT's share of consumer AI assistants dropped to 46.4% by late May — the first time under half. Google's Gemini climbed to 27.7% and Anthropic's Claude reached 10.3%, signaling a tightening race at the top of the assistant market.
China unveiled a $295 billion, five-year government AI infrastructure plan, about $59 billion a year in state-directed spend. Japan's Prime Minister Sanae Takaichi separately announced a plan to invest more than ¥370 trillion (roughly $2.3 trillion) through fiscal 2040, with ¥101.6 trillion earmarked specifically for AI and semiconductors.
Baseten, which runs AI models in production, closed a $1.5 billion Series F at a $13 billion valuation. The company says revenue grew about 20x year over year and it now handles more than one billion inference requests a day across 87 clusters and 18 clouds — evidence that capital is flowing toward inference compute, not just model training.
President Trump issued an executive order seeking more federal oversight of "frontier" AI models, asking companies to voluntarily share new models with the government for up to 30 days before wide release. It builds on the White House's National Policy Framework for AI, released March 20, which recommends Congress preempt state AI laws deemed overly burdensome in favor of a single national standard.
Noam Shazeer, a co-author of the seminal 2017 paper "Attention Is All You Need" that introduced the Transformer architecture, is leaving Google DeepMind to join OpenAI. The move is the latest high-profile defection in an intensifying talent war among the frontier labs.
On June 13, China's Zhipu AI released GLM-5.2 under a permissive MIT license. The launch continues the steady push of capable open-weight models that rival closed frontier systems while letting developers self-host and fine-tune freely, keeping pressure on proprietary providers.
Work presented at NeurIPS shows how inference-time scaling improves LLM reasoning on difficult problems by generating many solution attempts at once or exploring different reasoning paths, with a process reward model (PRM) scoring each to surface the most promising one. It is a core mechanism behind the current wave of reasoning models.
The frontier is crowded and fiercely competitive: Anthropic's Claude leads the benchmarks, Google's Gemini 3.5 Pro slips to July amid a researcher exodus, and ChatGPT's share dips below 50% for the first time. Money and policy are moving in lockstep — governments are committing trillions to sovereign AI compute, the US is tightening both export controls and frontier-model oversight, and investors are pouring capital into inference infrastructure like Baseten. Meanwhile open-weight models (GLM-5.2) and reasoning research (inference-time scaling) keep widening access and capability at once.