Builder map for local LLM lanes and Builds on named desk boxes. Open a lane or Build, then hop to Tok Lanes when you want the tok/s tables.

B70-class ≤32GB~55–70 tok/s on tuned ~35B MoE Q4 when it fits
Spark / Halo / Mac ~128GBcapacity for dense 70B+; Spark often ~3–8 tok/s batch-1
House 5080 16GB~12–15 tok/s when fit; <1 tok/s after spill

Featured

Lanes map

Local LLM lanes: speed, capacity, and the hybrid that isn’t

2026-09-09 · 5080 · B70 · Spark · Halo · Mac

One chart for fit and tok/s — who owns the ≤32GB speed lane vs the 128GB capacity lane.

Video companion: not run yet — placeholder only. Live channel: @llmlanes.

Builds

Lanes

Local LLM lanes: speed, capacity, and the hybrid that isn’t

2026-09-09 · 5080 · B70 · Spark · Halo · Mac

Result: Speed lane = B70-class ≤32GB; capacity = Spark/Halo/Mac ~128GB — not one hybrid SKU.
Compare

Quality tied — speed and VRAM decide

2026-09-08 · RTX 5080

Result: Gemma 12B ~70 tok/s vs Qwen 27B IQ4 ~28 tok/s — same 4/4 quality score.
MoE

FreeToken on WSL2 — mid MoE yes, Flash-Next no

2026-09-07 → 2026-09-08 · MoE · 5080

Result: 35B-A3B smoked (wall ~60–90); Flash-Next did not fit. Negative results count.
Quant

Same 27B, three quants — only one clears the bar

2026-09-08 · Unsloth · Ollama · 5080

Result: IQ4_XS full-GPU ~43 tok/s; library Q4 ~18; 46-layer spill ~7.
Editor

When the model “thinks,” the editor sees blank

2026-09-02 → 2026-09-08 · VS Code

Result: Answers landed in reasoning fields; think-off aliases fixed the UX.
Research

When the research swarm reports back

2026-09-09 · last30days

Result: One word (Intel), many public sources, one usable hardware map.

Method notes: Method notes · About: About · Tables: toklanes.com/reviews · Channel: @llmlanes.

Contact

[email protected]