Tok Lanes — measured tok/s from local runs

Numbers desk for measured tok/s from published LLM Lanes Builds. Compare tables here, then hop back to the builder map.

Measured runs — tok/s review bench

Gemma · IQ4 · FreeToken MoE · lanes map

Comparison tables pulled from published Builds. Cards link out to the full write-ups on llmlanes.com.

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

~70 tok/sGemma 4 12B · RTX 5080 · coding comparison · ~10.4GB VRAM

~28 tok/sQwen 27B Unsloth IQ4_XS · same tasks · ~15.6GB (tight)

~43 tok/s27B IQ4_XS full-GPU · Ollama ctx 4096 · vs ~18 / ~7 when spilled

~60–90 wallFreeToken 35B-A3B MoE on 16GB · Flash-Next did not fit

Single-stack readings restated from Builds — no new invented numbers. Tables: tok/s reviews. Write-ups: llmlanes.com/builds.

2026-09-21 · quality status

Result: B70 text all pass; 5080 mixed; vision describe 5080 PASS / B70 FAIL; gen BLOCKED.

~70 vs ~28 tok/s · same quality score

Result: Gemma 12B vs Qwen 27B IQ4 on identical coding tasks.

~43 tok/s full-GPU IQ4_XS · spills kill speed

Result: IQ4_XS stayed on-GPU; library Q4 and the layer recipe spilled.

35B-A3B yes · Flash-Next no

Result: Wall ~60–90 on short prompts when MoE fit; Flash-Next did not.

5080 · B70 · Spark · Halo · Mac tok/s chart

Result: Speed lane vs capacity lane — one chart for desk boxes.

Not a redirect to LLM Lanes — sister site with a tok/s review-bench focus.
Hard cross-links to the lanes map and Builds on llmlanes.com.
Numbers over hype — wall-clock and engine tok/s labeled as such when they differ.
About / method: About · method lives on llmlanes.com/measure; tables live here on tok/s reviews.

Sister map: LLM Lanes (site). Companion channel (live): @llmlanes. Tok channel (live): @toklanes. Open the Build · Compare tok/s on Tok Lanes.

[email protected]