Story cards stay below the sticky dock while audio, chapters, date, and brief navigation remain accessible.
A landmark study from Harvard Medical School published in Science found that OpenAI's o1 reasoning model correctly diagnosed 67% of emergency room patients using only text records, compared to 50-55% for triage physicians. Researchers stress this is not a deployment-ready finding but demonstrates an "urgent need for prospective trials" in real clinical settings.
The Department of Defense finalized agreements with OpenAI, Google, Microsoft, Nvidia, AWS, SpaceX, and Reflection AI to deploy AI on classified networks. Anthropic was blacklisted after refusing to allow the military to use Claude for "all lawful purposes" without safety guardrails — though a federal judge has since blocked the government's enforcement, and White House discussions have quietly resumed.
With its annualized revenue now past $30B and inference costs dragging margins, Anthropic is in early discussions with London-based Fractile about purchasing its SRAM-based inference accelerators that promise 25x faster inference at one-tenth the cost. Chips wouldn't be commercially ready until 2027, making this a medium-term diversification play alongside Nvidia, Google TPUs, and Amazon Trainium.
DeepSeek's V4 model is now available in two variants — V4-Pro (1.6T params, 49B active) and V4-Flash (284B total, 13B active) — both with 1M context. Promotional pricing through May 31 is an order of magnitude cheaper than competing frontier APIs, forcing a broad reassessment across the industry from chipmakers to LLM developers.
GPT-5.5 launched April 23 as OpenAI's most capable model, matching GPT-5.4's latency while performing at a higher level. OpenAI is positioning it as the backbone for agent-driven systems that execute complex multi-tool tasks. Available across ChatGPT, API, and Codex for Plus, Pro, Business, and Enterprise tiers.
Q1 earnings confirmed the hyperscalers are spending over $130B per quarter on AI infrastructure. Combined 2026 capex guidance now exceeds $725B — up from $410B in 2025. Market reaction was split: Alphabet and Amazon rose on cloud growth; Meta and Microsoft slipped as investors questioned ROI timelines.
Anthropic's run rate hit $30B in early April, up from $9B at end of 2025 — officially passing OpenAI's $25B. Over 1,000 enterprise customers now spend more than $1M/year on Claude. OpenAI disputes the comparison, claiming Anthropic's gross accounting overstates by ~$8B.
Masayoshi Son is preparing to list "Roze," a standalone AI and robotics company focused on building data centers with autonomous robots. It would bundle SoftBank's energy, infrastructure, and ABB Robotics assets. Target: $100B valuation, IPO in H2 2026 — though some executives consider the timeline ambitious.
Nvidia-backed Reflection AI — founded by former DeepMind researchers behind AlphaGo — secured a spot in the Pentagon's classified AI program while still valued at $25B without a publicly released model. The company raised $2B last year and plans a model trained on tens of trillions of tokens but has yet to deliver.
OpenAI's CFO Sarah Friar reportedly wants to delay the IPO from late 2026 to 2027, even as CEO Altman pushes forward. The company is valued at $852B after its $122B funding round but projects $14B in losses for 2026. OpenAI disputes the delay reporting, saying it hit Q1 revenue targets.