Subquadratic Launches SubQ With 12M-Token Context

Big claim, high upside, but the market will want independent validation before treating this as a real scaling breakthrough.

• –The core product is a long-context LLM, not just a research demo, with API and code-agent packaging already outlined.
• –The differentiation is architectural: SSA is meant to cut attention cost by avoiding full quadratic token-to-token processing.
• –The launch pitch is strong for developer workflows, especially repo-scale coding and retrieval across huge histories.
• –The downside is credibility risk: the headline benchmarks and efficiency claims are ambitious enough that third-party verification matters a lot.