Skip to content
AI News HubLIVE
More
Summary only

SubQ: Sub-quadratic LLM built for 12M-token reasoning

Subquadratic introduces SubQ, the first LLM built on a fully sub-quadratic sparse-attention architecture, enabling 12M-token reasoning with 150 tokens/s and 1/5 the cost of other leading models. It excels in long-context retrieval and coding benchmarks.

SourceHacker News AIAuthor: anujbans
SubQ: Sub-quadratic LLM built for 12M-token reasoning
Report an error

The correction channel is not available yet. You can copy the article reference below for later.

Correction instructions

Summary only

Under the source content permission, this page provides an editorial summary, key points, and a link to the original article.

Key points and analysis

Article intelligence

EngineersAdvanced

Key points

  • SubQ uses a sub-quadratic sparse-attention architecture, reducing attention compute by ~1000× at 12M tokens.
  • Supports 12M-token context for full repositories, long histories, and agent state.
  • Achieves 81.8% on SWE-Bench Verified, 95.0% on RULER @ 128K, and 65.9% on MRCR v2 (1M).
  • Available via API and Code products for developers and coding agents at 1/5 the cost.

Highlights and analysis are generated automatically and may contain errors. Check the original source.

More recent stories on this topic · Models