Alphabet has reportedly pushed the broader release of Gemini 3.5 Pro back by several months, after internal testing showed the system trailing the company's own targets on coding performance and complex, long-horizon reasoning tasks. The model, previewed at Google I/O and once expected around June, remains stuck in limited enterprise preview on Vertex AI.
It's the second missed deadline in a row — first the "give us until next month" promise at I/O, then a self-confirmed June 30 GA target that quietly slipped. Alphabet shares fell more than 4% following the reports, as investors weigh whether Google can hold its ground against Anthropic and OpenAI in the agentic-coding race CEO Sundar Pichai has already admitted the company is "a bit behind" on.
Speaking at a major tech summit in Shanghai, Chinese President Xi Jinping urged countries to cooperate on artificial intelligence and ensure that no single nation controls the technology — widely read as a direct jab at the United States. The remarks came as China showcased hardware and models it says rival the best coming out of Silicon Valley.
The new World AI Cooperation Organization (WAICO) is expected to give Beijing a formal platform to shape global AI norms and standards, positioning it against the voluntary-standards framework the White House has been pushing domestically. Analysts say the alliance is as much about geopolitical messaging as it is about actual coordination.
Apple has filed a lawsuit accusing OpenAI of stealing hardware trade secrets, escalating tension between the two companies just as both race to define what a dedicated AI consumer device looks like. The suit lands in the same 24-hour news cycle as a Chinese open-model release and a fresh EU order against Google — underscoring how crowded and adversarial the AI hardware and platform battle has become.
Neither company has detailed the specific IP at the center of the dispute, but the filing signals that the next phase of the AI arms race will be fought over silicon and device design, not just model weights.
Moonshot AI released Kimi K3, described as one of the largest open frontier models shipped to date, continuing a run of Chinese labs closing the gap with closed US systems on benchmarks while giving it away for free. The release lands in the same week Meta abandoned its own open-weight strategy in favor of a paid, closed model — a split that's reshaping who developers build on.
For builders watching the open-source lane, Kimi K3 adds to a crowded field that already includes DeepSeek and Alibaba's Qwen line, all competing on cost and permissive licensing rather than raw benchmark supremacy alone.
The Future of Life Institute's latest independent scorecard grades every major frontier lab on safety practice, transparency, and existential-risk planning — and the panel's headline finding isn't who's winning, it's that labs across the board are "quietly retreating from prior safety commitments" even as capabilities and deployment scale keep climbing.