Five AI stories from the week of May 4–May 10, 2026 that actually mattered — what changed, why it matters, and where to read more.
In this issue
- OpenAI makes GPT-5.5 Instant the default ChatGPT model
- Subquadratic launches SubQ with 12M-token context window and $29M seed
- NIST’s CAISI signs pre-deployment AI testing deals with Google DeepMind, Microsoft, xAI
- Perplexity opens Personal Computer agent to all Mac users
- Nvidia crosses $40B in AI equity deals in 2026, led by $30B OpenAI stake
OpenAI
OpenAI makes GPT-5.5 Instant the default ChatGPT model
On May 5, OpenAI replaced GPT-5.3 Instant with GPT-5.5 Instant as the default model across ChatGPT and the API (accessible as chat-latest), rolling out first to Plus and Pro subscribers. The update cuts hallucinated claims by 52.5% on high-stakes prompts in law, medicine, and finance, and adds context drawn from past conversations, connected files, and Gmail for personalized responses. Developers using chat-latest in the API will receive the new model automatically; GPT-5.3 remains selectable for paid users for three months before retirement.
Subquadratic
Subquadratic launches SubQ with 12M-token context window and $29M seed
Miami startup Subquadratic emerged from stealth on May 5 with SubQ, a large language model using sparse subquadratic attention that it says scales compute linearly rather than quadratically with context length, enabling a 12-million-token context window. The company raised $29M in seed funding and launched an API, a CLI coding agent (SubQ Code), and a long-context research search tool. Performance claims—including a purported 1,000× compute reduction over frontier models at full context length—have drawn skepticism from independent researchers who note no third-party reproduction exists yet.
Nextgov/FCW
NIST’s CAISI signs pre-deployment AI testing deals with Google DeepMind, Microsoft, xAI
On May 5, the Department of Commerce’s Center for AI Standards and Innovation announced voluntary pre-deployment testing agreements with Google DeepMind, Microsoft, and xAI, expanding a framework it established in 2024 with OpenAI and Anthropic. CAISI will evaluate models in classified environments—including versions with safety guardrails removed—to assess national-security-relevant capabilities before public release. The agreements are framed under the Trump administration’s AI Action Plan and remain voluntary, carrying no statutory enforcement mechanism.
TechCrunch
Perplexity opens Personal Computer agent to all Mac users
On May 7, Perplexity expanded its Personal Computer agent platform from a Max-subscriber waitlist to all Pro and Max subscribers through a new Mac desktop app. The agent combines a local runtime with more than 400 cloud connectors to execute multi-step workflows across files, native Mac apps, and the web, with all actions sandboxed and reversible; tasks can also be initiated remotely from an iPhone. Perplexity will deprecate its older Mac app as it consolidates around this product.
TechCrunch
Nvidia crosses $40B in AI equity deals in 2026, led by $30B OpenAI stake
As of May 9, Nvidia has committed more than $40 billion to equity positions in AI companies during 2026, anchored by a $30 billion stake in OpenAI agreed in February, with the remainder split across a $3.2 billion investment in fiber-optic supplier Corning, $2.1 billion in GPU-focused data-center operator IREN, seven other public-company stakes, and roughly two dozen private startup rounds. The pattern across deals is consistent: capital flows to companies that buy Nvidia GPUs at scale and re-rent compute to hyperscalers and model developers. Analysts have flagged circular-investment risk, noting Nvidia is effectively recapitalizing its own customers.