Note 003 · Architecture

The economics of hybrid attention

AUG 2026QWEN3.6-27B · GATED DELTANET · KV BUDGETSSTATUS IN PRODUCTION

TL;DR — Qwen3.6-27B is a hybrid: only every fourth block is full attention. The other 48 blocks are Gated-DeltaNet linear-attention blocks whose state is a fixed ~150 MiB — it does not grow with the conversation. So a token of context costs 34 KiB here versus the 136 KiB a dense 27B would pay: context is 4× cheaper by construction. That one architectural decision is what lets a single 24 GB GPU hold an 84K-token (~200-page) conversation next to the weights — and it is the quiet foundation under "free forever."

The stack, verified

The GGUF metadata says full_attention_interval = 4; the tensor names confirm it. Only blocks 3, 7, 11 … 63 carry attn_k / attn_v weights — 16 of 64 — plus block 64, the multi-token prediction head (Note 002). The remaining 48 blocks carry ssm_* recurrent-state tensors instead:

The Qwen3.6-27B stack drawn as 65 vertical slats: every fourth slat is a solid blue full-attention block (16 total), the thin slats between are Gated DeltaNet linear-attention blocks (48), and a hatched slat at the end is the multi-token prediction head.

FIG. 1 — One attention layer in four. Blue slats grow the KV cache; the thin ones hold a fixed-size recurrent state no matter how long you talk.

What a token of context costs

Per token, only the 16 attention blocks pay KV: 16 layers × (4 KV-heads × 256 key-length, keys and values) = 32,768 elements per token.

Cache typePer token262K-token cacheNote
f1664 KiB16.0 GiBUncompressed
q8_0 — chosen34 KiB8.5 GiBNear-lossless
q4_018 KiB4.5 GiBQuality degrades with depth
Dense 27B (hypothetical)136 KiB34.0 GiBAll 64 blocks attend
KV BYTES PER TOKEN OF CONTEXTSMALLER IS CHEAPER
Dense 27B136 KiB
Qwen3.6 hybrid34 KiB

Why q8, not q4

Halving the cache again with q4_0 would reach 128K context at full speed — but the quality cost grows with depth, which is exactly where long context matters:

ContextPerplexity, q8_0Perplexity, q4_0Gap
4K1.24231.2429Within noise
32K1.04161.0434~3× wider — and widening

The trend implies a wider gap still at 128K. q8_0 at 84K avoids the question entirely.

Where the constraint moved

Because context is 4× cheaper, the KV cache stopped being the binding constraint. The fixed costs dominate — 16.7 GiB of weights, 0.86 GiB of vision encoder — and the marginal cost of context, including the speculation head's draft buffers, measures ~44 MiB per 1K tokens at two slots. That is what makes the arithmetic of Note 001 close: weights + vision + 84K of q8 cache + margin ≈ 24 GB, with decode flat from 16K to 84K (50–57 tok/s) instead of decaying.

Why this makes free possible

The expensive part of serving long conversations is memory that grows with every token. A hybrid stack caps that growth at a quarter of the dense price, so one owned machine — no fleet, no per-token vendor bill — serves 200-page conversations at interactive speed. The architecture is the subsidy. There is no venture math hiding under the $0; there is a fixed-size recurrent state.