What comes after attention? This startup says it already knows.

Subquadratic is beginning to back up its ambitious claims with benchmarks and third-party validation for its SubQ 1.1 Small model, which uses its proprietary Sparse Attention (SSA) architecture to dramatically improve long-context performance. Rather than comparing every token to every other token, SSA selectively processes relationships, enabling near-linear scaling while maintaining high accuracy across context windows of up to 12 million tokens. The company reports near-perfect retrieval performance, competitive coding and reasoning benchmarks, and compute savings of up to 1,000x at maximum context lengths.

Rather than targeting frontier models immediately, Subquadratic is focusing on enterprise customers that need efficient analysis of massive datasets. The current model was built by replacing the dense attention mechanism in an existing open-weight model and then continuing long-context pretraining. Looking ahead, the startup plans to release a larger mid-tier model while continuing research into "zero attention" architectures that could eliminate attention mechanisms altogether, with the long-term goal of surpassing today's transformer-based AI models in both efficiency and capability.

Learn more from The New Stack around cloud spending:

The context window has been shattered: Subquadratic debuts a 12-million-token window

Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

Det här avsnittet är hämtat från ett öppet RSS-flöde och publiceras inte av Podme. Det kan innehålla reklam.

Avsnitt(300)

Why CPUs still matter in the age of AI agents

As AI evolves from conversational chatbots to autonomous agents, CPUs are becoming an increasingly important part of the infrastructure equation. In this episode, The New Stack speaks with Bhumik Pate...

11 Aug 26min

Why Doist Says Less AI Can Deliver More

Doist CTO Gonçalo Silva says AI is reshaping software development, but success depends on restraint rather than rapid feature expansion. Instead of chasing every AI capability, Doist prioritizes “subt...

31 Juli 35min

Why your company should (try to) build its own AI SRE

As AI coding agents accelerate software development, they also create new challenges for site reliability engineers (SREs), who are increasingly responsible for debugging systems that no single human ...

30 Juli 30min

Nvidia

In this episode with The New Stack Agents, Frederic Lardinois, NVIDIA’s Joey Conway says advances in AI over the past year have dramatically improved the capabilities of local models, making them prac...

23 Juli 32min

Meet Brain, the AI that decides when Azure is officially down

In this episode, Mark Russinovich, CTO of Microsoft Azure revealed Brain, the AI-powered AIOps system that continuously monitors Azure’s health, detects incidents, identifies root causes, and increasi...

14 Juli 19min

“The harness is where the hard work is”: Harness bets on agents that enterprises can trust in production

Harness has introduced Autonomous Worker Agents, a new capability that allows enterprises to replace rigid CI/CD pipeline scripts with AI agents that can deploy applications, run tests, and perform se...

2 Juli 19min

Public cloud vs. on-prem: Summit on where each workload belongs

More than two decades after AWS helped usher in the public cloud era, many organizations are reassessing whether a cloud-first strategy still delivers the cost and operational benefits it once promise...

25 Juni 36min