// NODE-02 · NEURAL FEED · DAILY TRANSMISSION //
April 28, 2026 · WINDYVIEWS.COM
OpenAI announced GPT-5.5, a substantial step up over GPT-5 on coding, computer use and deep research tasks. The model is rolling out to ChatGPT Plus, Pro, Business and Enterprise users, as well as inside the Codex IDE assistant. OpenAI is positioning the release as its most capable agent-grade model to date.
Chinese startup DeepSeek released a preview of its long-anticipated V4 open-source model family on April 24, claiming big gains in knowledge, reasoning and agentic tool use. V4 is optimized for popular agent harnesses including Anthropic's Claude Code and ships at significantly lower inference cost than prior versions. Analysts say the launch tightens an already close U.S.–China model race.
Microsoft and OpenAI amended their multi-year deal so OpenAI can serve its products on any cloud, not only Azure. In exchange, Microsoft drops the revenue-share it previously collected from OpenAI. The reset gives OpenAI freedom to court Google Cloud, AWS and Oracle while loosening one of the strongest ties in the modern AI stack.
Sony AI published Project Ace, the first known autonomous robot to play competitively against elite and professional table tennis players in the real world. The system combines high-speed perception, model-predictive control and a learned policy trained partly in simulation. Researchers describe it as the first time a robot has reached human-expert level in a popular physical sport.
At Google Cloud Next, Google introduced TPU 8t and TPU 8i, the first generation of TPUs to use two purpose-built dies — one tuned for training, the other for inference. Google says the split lets each chip optimize memory bandwidth and interconnect for very different workloads. The chips will power Gemini training and serving as well as third-party customers on Google Cloud.
The White House released its National Policy Framework for Artificial Intelligence on March 20 and continues lobbying Congress to enact it. The framework rejects a new federal AI agency, leans on existing regulators plus industry-led standards, and — most controversially — recommends federal preemption of state AI laws deemed to impose 'undue burdens.' Senator Marsha Blackburn's TRUMP AMERICA AI Act is the leading legislative vehicle.
Stanford HAI's 2026 AI Index shows Anthropic narrowly leading the frontier model rankings as of March 2026, with xAI, Google and OpenAI within striking distance and Chinese models like DeepSeek and Alibaba close behind. Top scores on SWE-bench Verified jumped from roughly 60% in 2024 to nearly 100% in 2025, while generative AI hit 53% population adoption inside three years — faster than the PC or the internet.
April 2026 is shaping up as the most consequential LLM month since the original GPT-4. Anthropic, Google, Meta, Alibaba and Mistral all shipped within two weeks, and the price of 'good enough' inference is roughly half what it cost in January 2026 — Sonnet 4 at \$3/\$15, Mistral Medium 3 at \$2/\$6, Gemini 2.5 Flash even cheaper. Buyers now have more viable model options than at any point in the field's history.
An updated version of AI Scientist, a system that autonomously generates hypotheses, designs experiments, runs them and writes the paper, had its first fully machine-authored submission accepted at a major machine-learning venue this month. The same April research wave includes SPPO, a new alignment technique that combines PPO-style training efficiency with more stable outcomes than standard RLHF. Both hint at how 2026 model recipes will look.
April 2026 is the inflection month: GPT-5.5, DeepSeek V4 and a wall of mid-tier releases dropped frontier and good-enough inference prices roughly in half, while Microsoft and OpenAI loosened their exclusive cloud bond. Hardware caught up — Google split its 8th-gen TPUs between training and inference — and Sony AI showed robotics is reaching human-expert level in real-world physical sport. Policy is sprinting to keep up: Washington is pushing a national framework that would preempt state AI laws, just as Stanford's 2026 AI Index confirms Chinese models are within touching distance of the U.S. frontier.