# SubQ

> **TL;DR.** SubQ is the brand name of the first large language model built on a fully sub-quadratic architecture, where attention compute scales linearly rather than quadratically with context length.

- **Category:** AI / Machine Learning / LLM Architecture
- **Stage:** validating
- **Age:** 42 days
- **Origin date:** 2026-05-05
- **First detected:** 2026-05-06
- **Canonical URL:** https://earlyterms.com/term/subq
- **Sources:** 6 primary URLs

## Definition

SubQ is the brand name of the first large language model built on a fully sub-quadratic architecture, where attention compute scales linearly rather than quadratically with context length. The model is developed by Miami-based startup Subquadratic using a proprietary mechanism called Subquadratic Sparse Attention (SSA).

Subquadratic launched SubQ on [May 5, 2026](https://subq.ai/introducing-subq) alongside a $29M seed round. At 12 million tokens, the company claims SSA reduces attention compute by roughly 1,000x versus dense-attention frontier models, with a 95% score on the RULER 128K benchmark at a cost of $8 — compared to $2,600 for Claude Opus 4.6 at the same benchmark.

## Analogy

Think of it as sparse indexing for databases — skip scanning every row, jump only to the relevant ones.

## Why it's emerging now

On May 5, 2026, Subquadratic launched SubQ with $29M seed funding and a single provocative claim: the first frontier LLM that ditches quadratic attention entirely. At 12 million tokens, SSA architecture cuts compute 1,000x vs. Claude Opus — but researchers are immediately demanding independent proof.

## Related terms

- *parent:* context-window
- *related:* context-engineering
- *related:* long-running-agents
- *related:* managed-agents
- *competitor:* claude-opus-4-6
- *competitor:* gpt-5-4
- *competitor:* gemini-3-1-pro
- *related:* Mamba
- *related:* RWKV
- *related:* FlashAttention
- *child:* Subquadratic Sparse Attention
- *child:* SubQ Code

## Sources

1. [Subquadratic — Introducing SubQ (official launch post)](https://subq.ai/introducing-subq)
2. [Subquadratic — How SSA Makes Long Context Practical](https://subq.ai/how-ssa-makes-long-context-practical)
3. [SiliconANGLE — Subquadratic launches with $29M to bring 12M-token context windows to AI](https://siliconangle.com/2026/05/05/subquadratic-launches-29m-bring-12m-token-context-windows-ai/)
4. [VentureBeat — Miami startup claims 1,000x AI efficiency gain; researchers demand independent proof](https://venturebeat.com/technology/miami-startup-subquadratic-claims-1-000x-ai-efficiency-gain-with-subq-model-researchers-demand-independent-proof)
5. [Hacker News — SubQ: Sub-quadratic LLM built for 12M-token context (discussion)](https://news.ycombinator.com/item?id=48023079)
6. [The Neuron Daily — SubQ ships 12M tokens at 1/5 the cost](https://www.theneurondaily.com/p/subq-ships-12m-tokens-at-1-5-the-cost)

---
_Generated by EarlyTerms · https://earlyterms.com/term/subq_