SubQ's model fits your project

PLUS: OpenAI's AI phone arriving early and Panthalassa's wave-powered ocean compute

Author

Gen AI Art
May 07, 2026 • Estimated Reading Time: 3 minutes

Sponsored by

A Miami startup just came out of stealth with a model that can hold 12 million tokens of context — your entire creative project, codebase, or script library — at a fraction of what frontier models charge for long context today.

That’s the pitch from Subquadratic’s SubQ, which uses a sub-quadratic architecture that scales cost linearly with input length instead of quadratically. If the benchmarks hold, a lot of the workarounds AI creators use to manage context limits — chunking, RAG, agent handoffs — could become unnecessary.

Today in AI:

  • SubQ’s 12M-token model launches with $29M in funding

  • OpenAI’s AI phone is arriving a full year early

  • Peter Thiel backs wave-powered AI data centers at sea

How Jennifer Aniston’s LolaVie brand grew sales 40% with CTV ads

The DTC beauty category is crowded. To break through, Jennifer Aniston’s brand LolaVie, worked with Roku Ads Manager to easily set up, test, and optimize CTV ad creatives. The campaign helped drive a big lift in sales and customer growth, helping LolaVie break through in the crowded beauty category.

What’s new? Subquadratic emerged from stealth with SubQ — a 12M-token LLM backed by $29M in seed funding that runs more than 50x faster and costs roughly 50x less than frontier models at 1 million tokens, using a novel sub-quadratic attention architecture.

What matters?

  • The SSA (Subquadratic Selective Attention) architecture scales linearly with input length, meaning cost stays manageable as context grows — a structural departure from the quadratic scaling that makes transformer long-context expensive.

  • On the RULER 128K benchmark, SubQ scored 95% accuracy at $8 — compared to ~$2,600 for Claude Opus at 94% accuracy — a roughly 300x cost reduction on the same task.

  • Two products launched today: a 12M-token API for developers and SubQ Code, a CLI coding agent that loads entire repositories in one pass without chunking or RAG scaffolding.

Why it matters?

For AI creators working on long-form projects — scripts, design briefs, multi-file codebases — a 12M-token window means loading everything into a single session without retrieval errors or context cutoffs. If the accuracy holds at scale, linear-cost long-context could significantly lower the barrier to using AI as a persistent creative collaborator.

GUIDE

What’s new? Supply chain analyst Ming-Chi Kuo reports that OpenAI is accelerating its first AI phone to mass production in the first half of 2027 — a full year ahead of previous estimates — with MediaTek as the exclusive chip supplier.

What matters?

  • The device will feature dual neural processing units running simultaneously — one for vision tasks, one for language — giving it persistent, real-time awareness of both what the camera sees and what the user says.

  • The phone includes an enhanced HDR pipeline in the image signal processor, signaling that visual AI perception — not just voice — is central to the device’s identity.

  • OpenAI’s IPO ambitions are likely accelerating the timeline, with combined 2027–2028 shipments projected at 30 million units if development stays on schedule.

Why it matters?

A phone designed from the ground up for simultaneous vision and language AI would fundamentally change how creators interact with AI tools — from text prompts to real-time visual understanding of their work environment. OpenAI is betting that the next AI interface isn’t a browser tab, it’s a camera.

SPONSORED BY MINTLIFY

Are you tracking agent views on your docs?

AI agents already outnumber human visitors to your docs — now you can track them.

What’s new? Panthalassa, an Oregon-based startup, raised $140M in Series B funding led by Peter Thiel to deploy autonomous 85-meter steel nodes in the Pacific Ocean — each converting wave energy into electricity for AI chips and cooled by seawater, with no land access or grid connection required.

What matters?

  • The floating nodes are fully self-sufficient: they generate power from wave motion, navigate autonomously using hull design, and transmit results via Starlink — sidestepping land permitting battles and power grid constraints that increasingly slow data center buildouts.

  • The funding will complete a Portland pilot factory and deploy the Ocean-3 pilot node series in the northern Pacific later in 2026, with commercial deployment targeted for 2027.

  • Additional investors include John Doerr, Marc Benioff’s TIME Ventures, and Max Levchin’s SciFi Ventures — signaling broad conviction in ocean-based compute infrastructure.

Why it matters?

Land-based data center expansion faces mounting community resistance and power grid limitations that are slowing AI infrastructure buildout in densely populated regions. Panthalassa’s bet on the ocean is creative in a literal sense — it’s a physical design solution to an increasingly political problem.

Everything else in AI

Anthropic launched ten ready-to-use AI agents for financial services, covering pitchbook creation, KYC screening, and earnings review — each bundled with domain-specific Claude models and Microsoft 365 connectors.

OpenAI released GPT-5.5 Instant as the new ChatGPT default, claiming 52.5% fewer hallucinated claims than its predecessor in high-stakes domains including medicine, law, and finance.

Cofounder released version 2, General Intelligence Company’s AI platform that lets founders orchestrate agents across engineering, sales, and marketing — running a full-stack startup without hiring anyone.

Essential AI Guides - Reading List:

Let us know!

Work with us

Reach 100k+ engaged Tech Professionals, Engineers, Managers and decision makers. Join brands like MorningBrew, HubSpot, Prezi, Nike, Ahref, Roku, 1440, Superhuman, and others in showcasing your product to our audience. Get in touch now →