2026-10-03 · 10 min
● CALIBRATION 001 · EVIDENCE-FIRST
What local AI actually runs on hardware you own.
Independent, reproducible local-AI benchmarks on real consumer hardware. Find what GPU, model, and quantization actually run on your rig.
§ 00 · QUICK ANSWERS
Start with the question you actually have
Straight answers grounded in our measured benchmark corpus — every number traceable to a reproducible report.
What should I expect?
Speed and quality by model size, quant, and hardware — measured, not vibes.
→ /what-to-expect/
Will it fit my GPU?
VRAM-to-model lookup: 8 to 24GB cards and 128GB unified-memory boxes.
→ /vram-lookup/
Mac, DGX Spark, or Strix Halo?
The unified-memory decision, with real tok/s from our Strix Halo bench.
→ /unified-memory/
Is the DGX Spark worth it?
A skeptical reality check on NVIDIA's desktop AI box — hype vs. evidence.
→ /dgx-spark-reality-check/
Which mini PC?
The Strix Halo tier list, anchored by our measured EVO-X2 data.
→ /mini-pcs/
Is a used GPU still worth it?
What the 3090 and friends actually deliver in 2026 — measured.
→ /used-gpus/
§ 01 · LATEST FINDING
From the bench
2026-10-03 · 12 min
The Private AI Stack for Small Teams: What to Actually Buy
2026-10-03 · 10 min
Two Cheap Boxes vs One Big Box: Clustering for Local AI
2026-09-18 · 11 min
Agentic coding on local models: which models survive the loop
§ 02 · EVIDENCE
Latest benchmark reports
All 0 reports →§ 03 · GO DEEPER