What's My AIA decision hub for the exact search queries people use before they benchmark, buy, or install.

decision guide

What local AI can I run?

The answer depends on which question you are asking. Use the benchmark when you have a machine in front of you, the model pages when you already have a target model, the runtime pages when setup flow is the blocker, and the hardware pages when you want evidence from similar shared machines.

indexed pages18
quarantined pages94
weekly batch20
priority routes6 P0

page map

Prioritized routes tied to real decision intent

These pages target the three concrete jobs people search for: model fit, hardware fit, and runtime fit. The general guide exists to route broad search demand into those deeper decision pages instead of chasing generic AI traffic.

P0Static

Guide

What local AI can I run?

Search intent: what local ai can i run

Top-level decision guide that routes people to the right model, runtime, or hardware page without generic AI fluff.

Benchmark product entry point

Open page
P0Static

Model page

Can I run gpt-oss-20b locally?

Search intent: gpt-oss-20b can i run it

34B class start • 15.5 GB minimum • 3 tracked runtime paths

Verified 2026-03-12 · review by 2026-04-11 • OpenAI

Open page
P0Static

Model page

Can I run Llama 3.3 70B locally?

Search intent: llama 3.3 70b can i run it

70B class start • 40.0 GB minimum • 3 tracked runtime paths

Verified 2026-03-12 · review by 2026-04-11 • Meta

Open page
P0Static

Model page

Can I run Phi-4-reasoning locally?

Search intent: phi-4-reasoning can i run it

13B class start • 8.5 GB minimum • 3 tracked runtime paths

Verified 2026-03-12 · review by 2026-04-11 • Microsoft

Open page
P0Static

Runtime page

Best local models for Ollama

Search intent: ollama best model

Best for the quickest path from benchmark result to a real local run.

Runtime guide + catalog coverage

Open page
P1Static

Model page

Can I run Llama 3.1 405B locally?

Search intent: llama 3.1 405b can i run it

Frontier MoE class start • 243.0 GB minimum • 3 tracked runtime paths

Verified 2026-03-12 · review by 2026-04-11 • Meta

Open page
P1Static

Model page

Can I run Llama 3.1 8B locally?

Search intent: llama 3.1 8b can i run it

7B class start • 6.5 GB minimum • 3 tracked runtime paths

Verified 2026-03-12 · review by 2026-04-11 • Meta

Open page
P1Static

Model page

Can I run Llama 4 Scout locally?

Search intent: llama 4 scout can i run it

120B class start • 67.0 GB minimum • 3 tracked runtime paths

Verified 2026-03-12 · review by 2026-04-11 • Meta

Open page

model intent

Model-name + “can I run it”

These pages stay grounded in the maintained catalog: minimum memory, practical hardware class, verified runtime coverage, and nearby model alternatives.

P0Static

Model page

Can I run gpt-oss-20b locally?

Search intent: gpt-oss-20b can i run it

34B class start • 15.5 GB minimum • 3 tracked runtime paths

Verified 2026-03-12 · review by 2026-04-11 • OpenAI

Open page
P0Static

Model page

Can I run Llama 3.3 70B locally?

Search intent: llama 3.3 70b can i run it

70B class start • 40.0 GB minimum • 3 tracked runtime paths

Verified 2026-03-12 · review by 2026-04-11 • Meta

Open page
P0Static

Model page

Can I run Phi-4-reasoning locally?

Search intent: phi-4-reasoning can i run it

13B class start • 8.5 GB minimum • 3 tracked runtime paths

Verified 2026-03-12 · review by 2026-04-11 • Microsoft

Open page

runtime intent

Runtime-name + “best model”

These pages answer the setup question directly: what to pull first in Ollama, LM Studio, or llama.cpp, and what tradeoff you are choosing with each runtime.

P0Static

Runtime page

Best local models for Ollama

Search intent: ollama best model

Best for the quickest path from benchmark result to a real local run.

Runtime guide + catalog coverage

Open page
P1Static

Runtime page

Best local models for llama.cpp

Search intent: llama.cpp best model

Best for people who care about low-level control, serving flags, and GGUF tuning.

Runtime guide + catalog coverage

Open page

hardware intent

Hardware-name + “what local AI”

Hardware intent now has two templates: static reference bands for common laptop and workstation questions, plus generated public-device pages when shared benchmark clusters are strong enough to trust.

Public shared-device pages are currently gated off. The reference hardware pages stay live so the site can ship high-intent hardware answers without pretending that sparse public proof already exists.

device intent

Named-device + “what local AI can I run”

These pages target the high-intent buyer who searches a named Mac, gaming laptop, workstation, or mini PC and wants a benchmark-backed reference answer instead of a generic buyer guide.

These templates stay in noindex quarantine until original benchmark evidence and manual reviewer promotion are recorded for the first publish batch.

gpu intent

GPU cohort + “what local AI”

GPU cohort pages answer the search pattern where buyers start with the accelerator tier itself and need a realistic model-size answer plus a direct path into the benchmark.

They remain quarantined from indexing until the template clears the people-first checklist and manual pass.

laptop families

Laptop family + memory band

Laptop family pages cover the buying language people actually use: MacBook Air, creator laptop, gaming laptop, workstation laptop, and similar family-level decisions.

They stay noindex until reviewer-approved benchmark evidence shows the family pages add distinct user value.

runtime clusters

Runtime + job-to-be-done pages

Runtime-intent clusters answer the mid-funnel search where people already chose Ollama, LM Studio, or llama.cpp and now need the best first model for a specific local AI job.

They remain in noindex quarantine until benchmark-backed proof shows the workflow pages outperform the canonical runtime guides.

P0Static

Runtime workflow

Best LM Studio models for a MacBook Pro

Search intent: best lm studio model macbook pro

34B class LM Studio page for Apple laptop owners who want the cleanest first desktop-app experience.

MacBook Pro workflow • launch week 1

Open page
P0Static

Runtime workflow

Best LM Studio models for beginners

Search intent: best lm studio model for beginners

13B class LM Studio page for the GUI-first audience that wants the shortest path to a first local win.

Beginner GUI workflow • launch week 1

Open page

template rules

How these pages scale without turning into spam

StaticTemplate rule

Generic “what local AI can I run” guide

  • Source: Homepage benchmark + page map
  • Publish when: Always. This is the hub that routes search traffic into benchmark, model, runtime, and hardware decisions.
  • Guardrail: Keep the benchmark CTA and the most useful decision links above the fold.
StaticTemplate rule

Model “can I run it” pages

  • Source: American model catalog + runtime coverage
  • Publish when: Only when a catalog entry has a maintained freshness window, minimum memory target, and at least one tracked local runtime path.
  • Guardrail: Use catalog-backed fit and runtime evidence only. Do not publish placeholder pages for untracked models.
StaticTemplate rule

Runtime “best model” pages

  • Source: Runtime registry + model catalog
  • Publish when: Only when the runtime has distinct setup tradeoffs and at least two tracked model fits worth comparing.
  • Guardrail: Rank models by fit and setup quality, not generic popularity or affiliate bias.
StaticTemplate rule

Reference hardware “what local AI can I run” pages

  • Source: Calibration reference bands + catalog fit
  • Publish when: Only when the reference band maps cleanly to a maintained model tier and the page links directly into benchmark and model/runtimes flows.
  • Guardrail: Keep these pages honest about being reference bands, not published shared-device proof.
GeneratedTemplate rule

Hardware fit pages

  • Source: Shared benchmark clusters + compare view
  • Publish when: Only when a device cluster has at least 3 shared runs and at least 55% deployment confidence.
  • Guardrail: Keep weak or sparse clusters out of the sitemap and link every indexed hardware page into compare and benchmark flows.

publishing plan

Weekly batches, review, and deindex rules

Week 120 pages

Reference device + GPU cohort + Laptop family + Runtime workflow

34B class gaming-laptop answer for a search cluster that usually means coding, chat, and first local tooling.

Example pages: What local AI can I run on a Lenovo Legion Pro 5 RTX 4060 laptop? • What local AI can I run on a Mac Studio M4 Max with 64 GB? • What local AI can I run on a MacBook Air M2 with 16 GB?

Week 220 pages

Reference device + GPU cohort + Laptop family + Runtime workflow

34B class creator-laptop answer for buyers who want style and local AI without carrying a thick gaming rig.

Example pages: What local AI can I run on a Dell XPS 14 RTX 4050 laptop? • What local AI can I run on a Framework 13 Ryzen AI laptop with 32 GB? • What local AI can I run on a Mac mini M4 with 16 GB?

Week 320 pages

Reference device + GPU cohort + Laptop family + Runtime workflow

34B class modular-laptop page for builders who want a repairable path into stronger local AI.

Example pages: What local AI can I run on a Framework 16 with Radeon 7700S? • What local AI can I run on a MacBook Air M3 with 24 GB? • What local AI can I run on a MacBook Pro M3 Max with 48 GB?

Week 420 pages

Reference device + GPU cohort + Laptop family + Runtime workflow

120B-capable mobile-workstation page for enterprise buyers who search named workstation lines before they buy.

Example pages: What local AI can I run on a Dell Precision RTX 5000 mobile workstation? • What local AI can I run on a home RTX 4070 Super tower with 64 GB? • What local AI can I run on a ThinkPad T14 Ryzen laptop with 32 GB?

Review20 pages/week

Quality review

  • Keep benchmark-first CTAs above the fold on every expansion page.
  • Noindex or consolidate pages that stop matching the search job cleanly.
  • Prefer catalog-backed claims and explicit tradeoffs over generic buying-guide copy.
Noindexunderperform or overlap

Deindex rules

  • Noindex thin parameter variants and duplicate filter states.
  • Consolidate overlapping cohorts into the strongest canonical landing page.
  • Drop pages that lose freshness, utility, or differentiated search intent.

next step

Start with the benchmark when you have the machine nearby.

Benchmark first for the fastest decision. Use the model and runtime pages when search starts with a model name or setup choice instead of a specific computer.