Moonshot’s Kimi K3 long-context design draws skepticism over 1M-token scaling
teortaxesTex · x · 2026-07-21
A thread questions whether Moonshot’s long-context research path is the right one, arguing that its post-R1 shift to MLA may be weak beyond 262K tokens and asking whether KDA can scale to 1M as well as CSA+HCA.
The attached architecture image shows Kimi K3’s design stack, including Stable LatentMoE, Gated MLA, KDA, and the Kimi Delta Attention backbone. The discussion frames long context as the AI era’s version of RAM and suggests the real moat may be organizational as much as model-level.
More from Models
- Fern's "context is intelligence" thesis: prediction keeps improving out to 10.2M tokens — bneyshabur · 2026-10-26
- Mirendil launches Fern, a from-scratch pretrained model with up to 10M-token context — bneyshabur · 2026-10-26
- Users spot a dedicated translation UI appearing in Claude — BLCNYY · 2026-10-26
- GrokBot user: flight booking and shopping all blocked, agent can't do anything useful — lucasmeijer · 2026-10-26
- Vending-bench-style eval: Sol 5.6 appears the most aggressive contract-violator — xuanalogue · 2026-10-26
- Fable 5.5 launch rumors persist for third week, but no frontier model drops this month — patience_cave · 2026-10-26