Moonshot open-sources MoBA, a block attention mechanism it says is 16x faster for long context
gekobraa · x · 2026-09-09
Moonshot AI (Kimi's maker) has open-sourced MoBA (Mixture of Block Attention), the attention mechanism behind Kimi's long context, claiming 95% of standard attention computation is wasted.
- Vanilla attention scales quadratically ($O(N^2)$): every token attends to all previous tokens
- MoBA applies Mixture-of-Experts ideas to the attention layer: context is chopped into blocks, and a lightweight dynamic gating lets each query token pick relevant blocks instead of brute-forcing full attention
- Reported up to 16x faster than full attention
More from Models
- Kimi K3.1 Pro tops MathArena open models, closing frontier gap to about 6 months — teortaxesTex · 2026-10-26
- Ruff creator Charlie Marsh: frontier models find infinite bugs in any software — charliermarsh · 2026-10-26
- Dev testing one skill across Opus 5.5 and Sol 6.1 finds instruction following 'all over the map' — lucasmeijer · 2026-10-26
- supermemory unveils learner-1, claiming continual learning for any model — julianweisser · 2026-10-26
- Anthropic says Opus 5.5 cuts cost per task by ~40% vs Opus 5 — taufiqintech · 2026-10-26
- AI search APIs ranked: top 12 within 7 points, yet prices differ nearly 7x — FinanceYF5 · 2026-10-26