gpt-oss 20B
3.6B active / 20B MoEnative MXFP4 reasoning model that stays practical on 16GB+ cards.
Quant
MXFP4 native
Speed
51.4-98.8
GPU layers
24/24
Score
92
Essential cookies keep the site working. Optional analytics are off unless you allow them.
Review cookie policy§ 001 / LOCAL LLM TOOLKIT
Tell it your GPU, workload, and tradeoff. It ranks 41 local models across 6 workloads and 3 priorities, then prefills VRAM checks before you download a 1590GB+ weight file.
§ 002 / PICKER
5/5 free runs left today
Current pick
native MXFP4 reasoning model that stays practical on 16GB+ cards.
Quant
MXFP4 native
Speed
51.4-98.8 tok/s
Ranked recommendations
native MXFP4 reasoning model that stays practical on 16GB+ cards.
Quant
MXFP4 native
Speed
51.4-98.8
GPU layers
24/24
Score
92
Qwen3.5 MoE quality with a small active path.
Quant
Q8_0
Speed
29.2-56.2
GPU layers
26/40
Score
91
multimodal MoE quality with a compact active path.
Quant
Q8_0
Speed
31.2-60
GPU layers
26/30
Score
91
current Qwen3.5 quality for 24GB-class cards.
Quant
Q8_0
Speed
11.4-22
GPU layers
53/64
Score
90
new Qwen3.8 multimodal model for 24GB-class cards.
Quant
Q8_0
Speed
11.4-22
GPU layers
53/64
Score
90
largest current dense Gemma 4 option.
Quant
Q8_0
Speed
9.9-19
GPU layers
44/60
Score
90
The full report for the top pick: recommended quant, the complete VRAM tradeoff table, and the llama.cpp launch command with the right --n-gpu-layers. Sent once, immediately.
Get the requested artifact now, then at most one evidence-backed Local AI Lab Note on Friday when there is something worth sharing. One-click unsubscribe. No sponsored placements. Privacy.
§ 003 / TOOL STATS
Default context
4K tokens
Scope
41 models / 6 workloads / 3 priorities
Recent usage
61 tracked runs / 30d
§ 004 / FAQ