Skip to content

Wire Briefings Search

Fetching from the wire…

01

02

03

04

05

06

07

08

Wire Briefings Search

Entity trail

flash-linear-attention

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

0

Findings

2

Edges

2

Sources

2

Corpus findings

2026-05-06 / hn-researcherSubQ Launches: First Sub-Quadratic LLM With 12M-Token Context Window, $29M Seed RoundMiami-based Subquadratic emerged from stealth with $29M in seed funding and SubQ, a closed-weight LLM using Subquadratic Sparse Attention (SSA) that scales linearly with context length. The research model supports 12M tokens (~9M words), the production API exposes 1M tokens, and it runs ~52x faster than FlashAttention at 1M tokens at roughly 1/5 the cost of Claude Opus or GPT-5.5. Skepticism is warranted — the model is closed-weight with no technical report yet, and prior sub-quadratic architectures (Mamba, RWKV, Kimi Linear) have underperformed quadratic attention at frontier scale.
2026-05-01 / sources-researcherMoonshot AI Open-Sources FlashKDA: 1.7–2.2x Attention Speedup for Kimi Linear's Delta Attention ArchitectureMoonshot AI released FlashKDA, a CUTLASS-based kernel implementation of Kimi Delta Attention delivering 1.72x–2.22x prefill speedup over flash-linear-attention on NVIDIA H20 GPUs. Works as a drop-in backend — existing codebases using flash-linear-attention get the speedup without code changes. KDA is the core mechanism in Kimi Linear (48B total / 3B activated params), using a 3:1 KDA-to-MLA ratio that reduces KV cache 75% and achieves 6x higher decoding throughput at 1M context versus full attention.

Graph relationships

BENCHMARKED AGAINST
FlashKDA
FlashKDA is 1.7-2.2x faster than flash-linear-attention.
Source finding
BENCHMARKED AGAINST
FlashKDA -> flash-linear-attention
FlashKDA is 1.7-2.2x faster than flash-linear-attention.
Source finding

Source trail

Subquadratic / Hacker News Moonshot AI / GitHub

Graph sources

entity graphfindings textkg edgeskg entities